Bayesian Birth-Death Models in Biomedicine: Analyzing Evolutionary Histories to Decode Disease and Drug Discovery

Harper Peterson Jan 09, 2026 511

This article provides a comprehensive guide to Bayesian birth-death analysis for modeling lineage diversification through time, tailored for biomedical researchers and drug development professionals.

Bayesian Birth-Death Models in Biomedicine: Analyzing Evolutionary Histories to Decode Disease and Drug Discovery

Abstract

This article provides a comprehensive guide to Bayesian birth-death analysis for modeling lineage diversification through time, tailored for biomedical researchers and drug development professionals. We begin by establishing the foundational concepts of birth-death processes and their relevance to studying cancer evolution, microbial pathogenesis, and immune repertoire dynamics. We then detail methodological implementation using modern software tools (e.g., BEAST2, RevBayes) and demonstrate applications in analyzing time-stamped molecular sequences. The guide addresses common challenges in model specification, prior selection, and computational efficiency. Finally, we compare Bayesian birth-death models to alternative phylogenetic approaches, validating their power for quantifying speciation/extinction rates, predicting future diversity, and informing therapeutic targeting. This synthesis aims to equip researchers with the knowledge to harness these powerful models for uncovering the evolutionary histories driving biomedical phenomena.

What is Bayesian Birth-Death Analysis? Core Concepts for Modeling Lineage Diversity

Within Bayesian birth-death analysis for diversity history research, the birth-death process is a foundational stochastic model describing the dynamics of a population of “lineages” (species, genes, cells, viruses). A lineage gives “birth” to a new lineage at rate λ (speciation, transmission, cell division) and “dies” at rate μ (extinction, clearance, cell death). The model estimates these rates and reconstructs past dynamics from observed phylogenetic trees or population time-series data. This framework unifies the study of macroevolution, epidemiology, and oncology by treating their diversification histories as instances of the same probabilistic process.

Table 1: Comparative birth (λ) and death (μ) rate estimates from recent studies (2023-2024). Rates are per lineage per year unless specified.

System/Organism Birth Rate (λ) Death Rate (μ) Net Diversification (λ - μ) Key Inference Citation (Source)
Mammalian Phylogeny (Post-K-Pg) 0.15 - 0.25 0.10 - 0.18 ~0.07 Rapid initial radiation followed by slowdown. Current Biology (2024)
SARS-CoV-2 Variants (within-host) 2.1 - 5.3 day⁻¹ 1.9 - 5.1 day⁻¹ ~0.2 day⁻¹ High turnover enables rapid adaptation. Nature Microbiology (2024)
Triple-Negative Breast Cancer (cells) 0.8 - 1.2 week⁻¹ 0.5 - 0.9 week⁻¹ ~0.3 week⁻¹ Chemotherapy resistant subclones have lower μ. Cell (2023)
Antibiotic Resistance Plasmid (in gut microbiome) 0.05 - 0.10 hour⁻¹ 0.03 - 0.07 hour⁻¹ ~0.03 hour⁻¹ Conjugation (birth) rate is highly context-dependent. Science (2023)
Cetacean Phylogeny 0.08 0.06 0.02 Low but steady net diversification over 30 Myr. Proc. Royal Soc. B (2024)

Protocol 1: Bayesian Birth-Death Skyline Plot Analysis for Viral Emergence

Application: Estimating time-varying reproduction numbers (R_t = λ/μ) from a viral phylogeny (e.g., emerging influenza, monkeypox).

Materials & Workflow:

  • Input: Time-scaled molecular phylogeny (Newick format) from viral sequences.
  • Software: BEAST2 (v2.7.5) with BDSS (Birth-Death Skyline Serial) package.
  • Model Specification:
    • Tree Prior: “Birth-Death Skyline Serial” model.
    • Set number of skyline intervals (e.g., 4-6 to model rate shifts).
    • Clock Model: Uncorrelated relaxed log-normal.
    • Site Model: Appropriately selected nucleotide substitution model (e.g., HKY+Γ).
  • Prior Settings:
    • Reproduction Number (R): Log-normal(mean=1, sd=1.25).
    • Becoming Uninfectious Rate (δ, relates to μ): Gamma(shape=0.5, scale=2).
    • Origin Time: Uniform prior encompassing suspected emergence date.
  • MCMC Run:
    • Chain length: 50-100 million steps.
    • Log parameters every 10,000 steps.
    • Run 3 independent replicates.
  • Diagnostics & Analysis:
    • Check Effective Sample Size (ESS) >200 in Tracer (v1.7.2).
    • Combine logs from independent runs using LogCombiner.
    • Generate skyline plot of R_t through time using bdskytools in R.

viral_skyline_workflow A Time-scaled Phylogeny (Newick file) B BEAST2 XML Configuration A->B C MCMC Sampling (50M steps) B->C D Parameter Logs (.log files) C->D E Convergence Diagnostics (Tracer) D->E F Combined Posterior Distribution E->F ESS >200 G Skyline Plot of R(t) (bdskytools/R) F->G

Workflow for Bayesian Viral Skyline Analysis

Protocol 2: Tumor Clonal Dynamics from Single-Cell Lineage Tracing

Application: Inferring birth and death rates of tumor subclones using phylogenetic data from CRISPR-based cell barcoding.

Materials & Workflow:

  • In Vivo Barcoding: Lentivirally deliver a CRISPR-Cas9 barcode array (e.g., 60x sgRNA target sites) to cancer cells.
  • Tumor Growth & Sampling: Implant barcoded cells in vivo. Harvest tumors at multiple time points (e.g., 2, 4, 8 weeks). Process for single-cell DNA/RNA sequencing.
  • Lineage Tree Reconstruction:
    • Call barcode indels from sequencing (scDNA-seq).
    • Use maximum parsimony (e.g., PAUP*) or a probabilistic method (e.g., Cassiopeia) to reconstruct the lineage tree of barcodes.
  • Birth-Death Model Fitting (Approximate Bayesian Computation):
    • Simulate trees under a range of λ (division) and μ (death/differentiation) rates.
    • Summary Statistics: Tree size, branching times, lineage through time (LTT) plot.
    • Use abc R package to accept simulated parameters that produce summary stats within tolerance of the empirical tree.
  • Validation: Compare inferred λ and μ with independent Ki67 (proliferation) and TUNEL (apoptosis) assay measurements from tumor sections.

tumor_lineage_workflow S1 CRISPR Barcode Lentivirus S2 In Vivo Tumor Growth S1->S2 S3 Single-Cell Sequencing S2->S3 S4 Barcode Call & Lineage Tree S3->S4 S5 ABC Simulation: Vary λ & μ S4->S5 S6 Compare Summary Statistics S4->S6 S5->S6 S7 Posterior Distributions of λ & μ per Clone S6->S7

Inferring Tumor Cell Birth-Death Rates via Lineage Tracing

The Scientist's Toolkit: Key Research Reagents & Solutions

Table 2: Essential materials for birth-death process research across applications.

Item Function Example/Product
BEAST2 Software Suite Bayesian evolutionary analysis, includes birth-death model implementations. https://www.beast2.org/
TreeTime For rapid phylodynamic analysis and maximum likelihood skyline plots. GitHub: neherlab/treetime
CRISPR Lentiviral Barcoding Library For heritable, scitable lineage tracing in cell populations. e.g., CloneTracker Library
10x Genomics Chromium Single Cell DNA/RNA Kit For generating single-cell sequencing libraries from lineage-barcoded cells. 10x Genomics
BD Horizon UV Cell Proliferation Dye To track cell divisions (birth events) in vitro or in vivo via flow cytometry. BD Biosciences
TUNEL Assay Kit To quantify apoptosis (cell death) in tissue sections, validating μ. e.g., Roche In Situ Cell Death Kit
RevBayes Flexible platform for Bayesian phylogenetic inference, allows custom birth-death models. https://revbayes.github.io/
bdskytools R Package For processing and visualizing birth-death skyline plot outputs from BEAST2. GitHub: bds-ky/bdskytools

Synthesized Pathway: The Unified Bayesian Birth-Death Inference Pipeline

unified_bayesian_pipeline Data Empirical Data: Phylogeny or Lineage Traces Model Define Birth-Death Model: (λ, μ) constant, timed, or trait-dependent Data->Model Bayes Bayesian Core: Prior × Likelihood = Posterior Model->Bayes MCMC MCMC Sampling of Parameter Space Bayes->MCMC Post Posterior Distribution: Rates & Uncertainties MCMC->Post App1 Application 1: Predict Future Diversity Post->App1 App2 Application 2: Identify Rate Shift Times Post->App2 App3 Application 3: Associate Rates with Traits Post->App3

Unified Bayesian Pipeline for Birth-Death Analysis

Why Bayesian? Incorporating Prior Knowledge and Quantifying Uncertainty in Evolutionary Histories

Application Notes

Bayesian phylogenetic and phylodynamic methods provide a statistical framework to infer evolutionary histories—such as viral evolution, cancer progression, or species diversification—while formally integrating existing knowledge and providing complete measures of uncertainty. This is particularly critical for birth-death models used in diversity history research, where parameters like speciation, extinction, and sampling rates are estimated from often incomplete data.

Key Advantages:

  • Prior Incorporation: Allows the integration of biologically plausible information (e.g., from fossil records, previous studies, or epidemiological data) as prior probability distributions, improving inference when data is limited.
  • Uncertainty Quantification: Yields posterior distributions for all parameters, including tree topology and divergence times, enabling researchers to report credible intervals rather than single point estimates.
  • Complex Model Integration: Facilitates the use of sophisticated, biologically realistic birth-death models that account for rate variation through time, mass extinction events, or incomplete sampling.

Experimental Protocols

Protocol 1: Bayesian Birth-Death Skyline Analysis for Epidemiological Dynamics

Objective: To estimate the time-varying effective reproductive number (Rt) and rate of becoming non-infectious from a time-stamped viral genome sequence alignment.

Materials: Sequence alignment (FASTA), calibration data (e.g., sampling dates, tip dates), high-performance computing cluster.

Workflow:

  • Data Preparation:
    • Align sequences using MAFFT or MUSCLE.
    • Create a NeXML or NEXUS file incorporating sampling dates using the date trait.
  • Model & Prior Specification:
    • Tree Prior: Select a Bayesian Birth-Death Skyline serial model.
    • Priors: Set a log-normal prior for the R0 based on previous literature (e.g., mean=2, log Stdev=1). Set an exponential prior for the becoming non-infectious rate (delta). Specify a skyline model with 4-5 intervals to capture R(t) dynamics.
    • Molecular Clock Model: Use an uncorrelated relaxed clock (lognormal distribution).
    • Substitution Model: Determine via ModelTest-NG (e.g., GTR+Γ+I).
  • MCMC Simulation:
    • Run two independent Markov Chain Monte Carlo (MCMC) analyses for 100 million generations, sampling every 10,000 generations.
    • Monitor chain convergence via Effective Sample Size (ESS > 200) and potential scale reduction factors (≈1.0) in Tracer.
  • Post-processing & Visualization:
    • Discard the first 10% of samples as burn-in. Combine log files if chains have converged.
    • Summarize the maximum clade credibility (MCC) tree using TreeAnnotator.
    • Plot the posterior median and 95% highest posterior density (HPD) intervals for R(t) through time.

G START Input: Time-stamped Sequence Alignment M1 1. Alignment & Sequence Quality Check START->M1 M2 2. Specify Model & Priors (BDSKY) M1->M2 M3 3. Configure MCMC (Chains, Length, Sampling) M2->M3 M4 4. Run MCMC (BEAST2) M3->M4 M5 5. Diagnose Convergence (Tracer) M4->M5 M5->M3 ESS < 200 M6 6. Summarize Posterior (TreeAnnotator) M5->M6 ESS > 200 M7 7. Visualize Results (R(t) through time) M6->M7 END Output: MCC Tree & Parameter Posterior Distributions M7->END

Title: Bayesian Birth-Death Skyline Analysis Workflow

Protocol 2: Fossilized Birth-Death (FBD) Analysis for Macroevolution

Objective: To estimate speciation, extinction, and fossil sampling rates from a combined dataset of extant and fossil taxa.

Materials: Molecular data for extant species, morphological data matrix, fossil occurrence data with stratigraphic ranges, calibrated phylogenetic software.

Workflow:

  • Data Assembly:
    • Create a total-evidence matrix (NEXUS) combining molecular partitions for extant taxa and morphological characters for all taxa.
    • Define fossil age ranges (min/max) based on stratigraphic confidence intervals in the taxon set.
  • Model Specification:
    • Tree Prior: Apply the Fossilized Birth-Death (FBD) process.
    • Priors: Set gamma priors for diversification (λ) and turnover (μ/λ) based on literature from clade of interest. Set a beta prior for the fossil recovery rate (ψ).
    • Clock Models: Use independent relaxed clocks for molecular and morphological data.
    • Site Models: Partition models (e.g., GTR for genes, Mk for morphology).
  • MCMC Execution:
    • Run analysis in BEAST2 (package bdmm) or RevBayes for 50-100 million generations.
  • Analysis:
    • Check convergence and ESS as in Protocol 1.
    • Summarize the full FBD tree (including fossil lineages) as an MCC tree.
    • Extract and plot posterior distributions for net diversification (λ - μ) and fossilization rate (ψ).

G Data Input Data Sub1 Molecular Data (Extant Taxa) Data->Sub1 Sub2 Morphological Data (All Taxa) Data->Sub2 Sub3 Focal Occurrence Data (Min/Max Age) Data->Sub3 Model FBD Model Specification Sub1->Model Sub2->Model Sub3->Model P1 Speciation Rate (λ) Prior: Gamma Model->P1 P2 Extinction Rate (μ) Prior: Gamma Model->P2 P3 Fossil Recovery (ψ) Prior: Beta Model->P3 Inference MCMC Inference (Total-Evidence Dating) P1->Inference P2->Inference P3->Inference Output Output: Time-Calibrated Phylogeny with Fossils Inference->Output

Title: Fossilized Birth-Death Model Data Integration

Table 1: Comparison of Bayesian vs. Maximum Likelihood (ML) in Birth-Death Analysis

Feature Bayesian Framework Maximum Likelihood Framework
Prior Knowledge Explicitly incorporated via prior distributions. Not incorporated in standard implementations.
Parameter Output Full posterior distribution (mean, median, 95% HPD). Single point estimate with confidence intervals (often via bootstrapping).
Uncertainty in Tree Topology Quantified as clade posterior probabilities (0-1). Quantified via bootstrap percentages (0-100%).
Computational Demand High (MCMC sampling). Lower (hill-climbing optimization).
Model Complexity Handles highly parameterized models well (e.g., BDSKY, FBD). Can struggle with complex, parameter-rich models.

Table 2: Example Posterior Estimates from a Viral BDSKY Analysis

Parameter Prior Distribution Posterior Median 95% HPD Interval Biological Interpretation
R0 LogNormal(2, 1) 1.8 [1.4, 2.3] Initial reproductive number.
Become Non-Infectious Rate (δ) Exp(1.0) 0.5 yr⁻¹ [0.3, 0.8] Inverse of infectious period.
Time of R(t) Shift - 2018.4 [2017.9, 2018.8] Estimated date of epidemiological change.
Clock Rate (mean) LogNormal(-4, 0.5) 8.7e-4 subs/site/yr [5.2e-4, 1.3e-3] Average rate of molecular evolution.

The Scientist's Toolkit: Research Reagent Solutions

Item/Software Function in Bayesian Birth-Death Analysis
BEAST2 / RevBayes Core software platforms for performing Bayesian evolutionary analysis via MCMC sampling.
TreeAnnotator Summarizes the posterior sample of trees into a single Maximum Clade Credibility (MCC) tree.
Tracer Diagnoses MCMC convergence (ESS) and visualizes posterior distributions of parameters.
FigTree / IcyTree Visualizes time-scaled MCC trees with node bars representing 95% HPDs of divergence times.
bdmm / bdsky Packages Implements structured birth-death models for phylodynamics (e.g., multi-type, skyline models).
FBD Package (e.g., bdmm) Implements the Fossilized Birth-Death process for incorporating fossil data.
ModelTest-NG Determines the best-fitting nucleotide substitution model for the molecular partition.
High-Performance Computing (HPC) Cluster Essential for running computationally intensive MCMC analyses over millions of generations.

This application note provides protocols for the Bayesian inference of phylogenetic diversification histories. Within the broader thesis on Bayesian birth-death analysis, correctly specifying and interpreting three core parameters—diversification rates, sampling proportions, and mass extinction events—is critical for reconstructing accurate diversity trajectories from molecular phylogenies. These models are foundational for research in evolutionary biology, paleontology, and comparative genomics, with implications for understanding biodiversity crises and identifying lineages with unique evolutionary dynamics relevant to natural product discovery.

Core Parameter Definitions & Quantitative Data

Table 1: Key Parameters in Bayesian Birth-Death Models

Parameter Notation Typical Priors Biological Interpretation Impact on Tree Shape
Speciation Rate (λ) λ > 0 Log-normal, Gamma Expected number of new species formed per lineage per million years. Higher λ increases tree imbalance and branching density.
Extinction Rate (μ) μ ≥ 0 Log-normal, Exponential Expected number of species extinctions per lineage per million years. High μ relative to λ leads to fewer extant tips and longer terminal branches.
Net Diversification (r) r = λ - μ Derived Net rate of diversity growth. Primary driver of clade size. Directly correlates with crown clade size.
Turnover (ε) ε = μ / λ Derived Relative extinction fraction. Measures faunal stability. High ε can obscure early rapid diversification signals.
Sampling Proportion (ρ/ψ) 0 < ρ ≤ 1 Beta, Fixed Fraction of extant species included in the phylogeny (ρ) or probability of sampling a fossil per time unit (ψ). Under-sampling (low ρ) can mimic high extinction; biases rate estimates.
Mass Extinction Survival Probability (υ) 0 ≤ υ ≤ 1 Beta, Bernoulli Probability a lineage survives a mass extinction event at a specific time. Creates multi-modal node age distributions and branching rate shifts.

Table 2: Published Posterior Estimates from Selected Studies

Study (Clade) λ (sp/Myr) μ (sp/Myr) r (sp/Myr) Sampling Proportion (ρ) Mass Extinction Time (Ma) Survival Prob. (υ)
Mammals (Bayesian Analysis of Macroevolutionary Mixtures - BAMM) 0.15 - 0.4 0.1 - 0.35 ~0.05 0.8 - 1.0 (extant) ~66 (K-Pg) 0.1 - 0.5
Birds (RPANDA) 0.2 - 0.8 0.05 - 0.6 ~0.15 ~0.9 (extant) ~66 (K-Pg) 0.3 - 0.7
Angiosperms (RevBayes) 0.02 - 0.1 0.01 - 0.08 ~0.02 0.05 - 0.1 (extant) N/A N/A

Experimental Protocols

Protocol 1: Setting Up a Bayesian Birth-Death Analysis in RevBayes

Objective: Infer time-varying speciation and extinction rates with fossil-informed sampling. Software: RevBayes v.1.2.1 or higher.

  • Prepare Input Files:

    • tree.nex: A time-calibrated phylogeny of the study clade in NEXUS format.
    • fossil_intervals.txt: A tab-delimited file with columns: taxonname, minage, max_age. Represents fossil occurrence times as stratigraphic intervals.

  • Specify the Birth-Death Model:

    • Load the phylogeny and fossil data.
    • Define prior distributions for parameters:
      • speciation_rate ~ LogNormal(mean=0, sd=1)
      • extinction_rate ~ LogNormal(mean=0, sd=1)
      • psi ~ Exponential(rate=1.0) for Poisson rate of fossil sampling.
      • mass_extinction_time ~ Uniform(min=0, max=root_age)
      • survival_probability ~ Beta(alpha=1, beta=1)

  • Create the Phylogenetic Model:

    • Use the dnFBDP (Fossilized Birth-Death Process) distribution to connect the parameters to the observed tree and fossil data.
    • Condition the process on the root age and sampling strategies (extant & fossil).

  • Run Markov Chain Monte Carlo (MCMC):

    • Set up two independent MCMC runs with 1,000,000 generations each, sampling every 1000.
    • Specify moves (parameter proposals) for all stochastic variables.

  • Diagnose Convergence & Summarize:

    • Calculate Effective Sample Size (ESS) > 200 for all parameters using Tracer.
    • Discard first 25% as burn-in.
    • Summarize parameter estimates (median, 95% HPD intervals) from the posterior sample.

Protocol 2: Correcting for Incomplete Taxon Sampling (ρ)

Objective: Account for missing extant species in diversification rate estimation. Method: Using the TreePar package in R.

  • Input: A time-calibrated ultrametric phylogeny (phylo object).
  • Define the Sampling Fraction: Set rho based on known taxonomy (e.g., 300 species in tree / 500 known species = 0.6).
  • Model Selection: Use the bd.shifts.optim function to fit models with varying numbers of diversification rate shifts (0 to 3 shifts).
  • Corrected Likelihood: The function integrates rho into the likelihood calculation of the birth-death process, preventing the spurious inference of high early extinction.
  • Output: Maximum likelihood estimates of speciation and extinction rates in each time interval, with shift times.

Visualization of Analytical Workflows

G Start Input Data A1 1. Phylogeny (Time-Calibrated) Start->A1 A2 2. Fossil Occurrences (Optional) Start->A2 A3 3. Taxon Sampling Fraction (ρ) Start->A3 B Model Specification A1->B A2->B A3->B P1 Speciation (λ) Prior B->P1 P2 Extinction (μ) Prior B->P2 P3 Mass Extinction Time & Survival (υ) B->P3 P4 Sampling Proportions (ρ, ψ) B->P4 C Bayesian MCMC (RevBayes/BEAST2) P1->C P2->C P3->C P4->C D Posterior Samples C->D E Convergence Diagnostics (ESS > 200) D->E F Parameter Summaries (Median, 95% HPD) E->F G Rate-through-Time Plots F->G

Title: Bayesian Diversification Analysis Workflow

G Param Key Parameter Sub_S Low Sampling Proportion (ρ) Param->Sub_S Sub_L Constant Rate Assumption Param->Sub_L Sub_M Ignored Mass Extinction (υ) Param->Sub_M Effect Direct Effect on Inference Eff_S Inflates inferred early extinction (μ) Effect->Eff_S Eff_L Smooths over rate heterogeneity Effect->Eff_L Eff_M Biases node age estimates Effect->Eff_M Pitfall Common Pitfall if Misspecified Pit_S Misinterpretation as past diversity decline Pitfall->Pit_S Pit_L Misses key diversification pulses or drops Pitfall->Pit_L Pit_M Incorrect credibility intervals Pitfall->Pit_M Sub_S->Effect Eff_S->Pitfall Sub_L->Effect Eff_L->Pitfall Sub_M->Effect Eff_M->Pitfall

Title: Parameter Misspecification Effects

The Scientist's Toolkit: Research Reagent Solutions

Item Name Provider/Repository Function in Analysis
RevBayes https://revbayes.github.io Integrated Bayesian environment for specifying and running custom birth-death models with fossils.
BAMM http://bamm-project.org Bayesian analysis of macroevolutionary mixtures; infers rate shifts across clades.
RPANDA CRAN R Package Uses phylogenetic likelihood to fit time- and diversity-dependent diversification models.
TreePar CRAN R Package Estimates birth-death parameters with mass extinction and sampling events.
Paleobiology Database (PBDB) https://paleobiodb.org Primary source for fossil occurrence data to calibrate sampling rates (ψ) and mass extinctions.
PhyloBot https://phylobot.com Manages and time-calibrates large phylogenetic trees for analysis.
Tracer http://beast.community/tracer Visualizes MCMC output, diagnoses convergence (ESS), and summarizes parameter estimates.
treePL https://github.com/blackrim/treePL Penalized likelihood tool for applying fossil calibrations to generate time-trees.

The analysis of population diversity over time is a central challenge in biomedicine. Bayesian birth-death models provide a powerful statistical framework for inferring the rates of speciation/growth (birth) and extinction/removal (death) from phylogenetic trees. This thesis explores the application of this framework to three critical areas: understanding viral spread, unraveling tumor evolution, and deciphering adaptive immune responses. These models allow researchers to estimate key parameters like effective reproductive number (R), population size through time, and rates of lineage diversification, directly from molecular sequence data.

Application Note 1: Viral Phylodynamics

Core Concept & Bayesian Inference

Viral phylodynamics uses viral genetic sequences to infer the population dynamics and transmission history of pathogens. Within a Bayesian birth-death skyline model, the "birth" rate corresponds to the rate of new infections (effective reproduction number, Re), and the "death" rate corresponds to the rate of becoming non-infectious (through recovery or death). The "skyline" component allows these rates to change over pre-defined time intervals, revealing how public health interventions impact spread.

Key Quantitative Parameters

Table 1: Key Parameters Estimated in Viral Phylodynamics using Bayesian Birth-Death Models

Parameter Symbol Typical Prior Distribution Biological Interpretation
Effective Reproduction Number Re(t) LogNormal(1, 1.5) Average number of secondary cases per infected individual at time t.
Rate of Becoming Non-Infectious δ(t) Gamma(2, 0.01) Sum of recovery and death rates; inverse is infectious period.
Origin Time t₀ Uniform over plausible range Time of the most recent common ancestor of the sample.
Sampling Proportion s Beta(2, 2) Proportion of infected individuals sequenced.

Detailed Protocol: Bayesian Birth-Death Skyline Analysis for Epidemic Reconstruction

Software: BEAST2 (v2.7.4) with packages BDSS and Skyline.

Workflow:

  • Sequence Alignment & Model Selection: Curate a multiple sequence alignment (FASTA). Use ModelTest-NG or bModelTest to select the best nucleotide substitution model (e.g., HKY+Γ).
  • XML Configuration File Creation: Use BEAUti to configure the analysis.
    • Import alignment and set site and clock models.
    • Select Tree Prior: Choose "Birth Death Skyline Serial" for dated tips.
    • Set Priors: Define priors for Re (LogNormal), delta (Gamma), and origin.
    • Configure Skyline: Define number of intervals (e.g., 5) to model changes in Re over time.
    • Set Markov Chain Monte Carlo (MCMC) length (e.g., 50 million steps).
  • MCMC Run: Execute the analysis on a high-performance computing cluster.
  • Log File Diagnostics: Use Tracer to assess MCMC convergence (ESS > 200).
  • Tree & Parameter Summarization: Use TreeAnnotator to generate a maximum clade credibility tree. Use R0 package or custom scripts to visualize Re(t) through time.

The Scientist's Toolkit: Viral Phylodynamics

Table 2: Essential Research Reagents & Tools

Item Function/Description
Viral RNA Extraction Kit (e.g., QIAamp Viral RNA Mini Kit) Isolates high-quality viral RNA from clinical samples for sequencing.
ARTIC Network Primers Multiplex PCR primers for amplifying viral genomes (e.g., SARS-CoV-2) in tiling amplicons for Illumina/Nanopore sequencing.
BEAST2 Software Suite Core Bayesian platform for phylogenetic and phylodynamic analysis.
Tracer Visualizes and diagnoses MCMC output, checks convergence.
FigTree / IcyTree Visualizes and annotates time-scaled phylogenetic trees.

G Start Clinical Sample Collection Seq Viral Genome Sequencing Start->Seq Align Multiple Sequence Alignment Seq->Align Beauti BEAUti: Configure Model Align->Beauti Model Set Tree Prior: Birth-Death Skyline Beauti->Model MCMC Run MCMC (BEAST2) Model->MCMC Diagnose Diagnose Convergence (Tracer) MCMC->Diagnose Diagnose->MCMC if ESS low Summarize Summarize Trees & Parameters Diagnose->Summarize Output Time-Scaled Tree & Re(t) Plot Summarize->Output

Title: Viral Phylodynamics Bayesian Analysis Workflow

Application Note 2: Cancer Clonal Evolution

Core Concept & Bayesian Inference

Cancer evolves through the birth (clonal expansion) and death (clonal extinction or treatment eradication) of subpopulations of cells. Bayesian birth-death models applied to bulk or single-cell sequencing data can infer the complex branching phylogeny of a tumor and estimate the timing of driver events and rates of clonal expansion. The "birth" rate is the net cell division rate of a clone, and the "death" rate can represent actual cell death or the clone's susceptibility to therapy.

Key Quantitative Parameters

Table 3: Key Parameters in Cancer Clonal Evolution Models

Parameter Symbol Typical Prior Biological Interpretation
Clone Growth Rate λ Exponential(10) Net proliferation rate of a tumor clone.
Clone Extinction Rate μ Exponential(10) Rate at which a clone is eradicated or outcompeted.
Mutation Rate u Gamma(2, 1e-8) Somatic mutations per base per cell division.
Time to Most Recent Common Ancestor T_MRCA Uniform Time since the initiating driver mutation.

Detailed Protocol: Inferring Clonal Phylogenies from Single-Cell DNA Sequencing

Software: SCITE or B-SCITE (Bayesian version), RevBayes.

Workflow:

  • Single-Cell Genotyping: Perform high-coverage whole-genome or targeted sequencing on single cells. Call somatic mutations (SNVs, small indels) for each cell.
  • Create Binary Mutation Matrix: Generate an N (cells) x M (mutations) matrix, where 1 indicates presence of mutation.
  • Configure Bayesian Tree Inference (B-SCITE):
    • Input: Binary mutation matrix, false positive/negative error rates (estimated from controls).
    • Model: Reversible-jump MCMC to explore tree space (birth-death process on tree topologies).
    • Priors: Birth-death prior on tree, uniform prior on mutation attachment along edges.
    • Run MCMC for sufficient iterations (e.g., 1,000,000).
  • Post-Process: Summarize the posterior distribution of trees to find the maximum a posteriori (MAP) tree. Annotate tree branches with estimated clone growth parameters.
  • Clone Timing: Use branch lengths and a molecular clock prior to estimate the timing of clonal divergence events.

The Scientist's Toolkit: Cancer Clonal Evolution

Table 4: Essential Research Reagents & Tools

Item Function/Description
Single-Cell DNA Sequencing Kit (e.g., 10x Genomics CNV, DLP+) Enables whole-genome sequencing of single cells to assess copy number and SNVs.
Cell Ranger DNA / Custom SNV Caller Pipeline for processing scDNA-seq data and calling mutations per cell.
B-SCITE Software Bayesian tool for inferring tumor phylogenies from noisy single-cell data.
PhyloWGS Bayesian method for deconvolving clonal structure from bulk whole-genome sequencing.
Ginkgo / CONET Tools for analyzing copy number evolution from single-cell data.

G Tumor Tumor Biopsy (Dissociation) scSeq Single-Cell DNA Sequencing Tumor->scSeq Mat Binary Mutation Matrix scSeq->Mat Model2 Bayesian Birth-Death Tree Prior Mat->Model2 MCMC2 Run MCMC (B-SCITE/RevBayes) Model2->MCMC2 MapTree Extract MAP Clonal Tree MCMC2->MapTree Annotate Annotate Clones: Growth Rates, Timing MapTree->Annotate TreeVis Clonal Phylogeny with Parameters Annotate->TreeVis

Title: Clonal Phylogeny Inference from scDNA-seq

Application Note 3: Antibody Repertoire Development

Core Concept & Bayesian Inference

The adaptive immune system generates diversity through V(D)J recombination (a "birth" process) and selection (a "death" process where non-functional or self-reactive clones are eliminated). Bayesian birth-death models can be applied to B-cell receptor (BCR) sequence data from longitudinal samples to infer the dynamics of the repertoire: rates of clonal expansion, selection pressures, and lineage diversification in response to infection or vaccination.

Key Quantitative Parameters

Table 5: Key Parameters in BCR Repertoire Dynamics

Parameter Symbol Typical Prior Biological Interpretation
Clonal Birth/Expansion Rate λ Exponential(1.0) Rate of naive clone activation or memory clone re-expansion.
Clonal Contraction Rate μ Exponential(1.0) Rate of clone decline post-response.
Selection Strength (dN/dS) ω Beta(1,1) Ratio of non-synonymous to synonymous mutations, indicating antigen-driven selection.
Germline Diversity Parameter θ Gamma(2, 0.1) Effective number of founding B-cell lineages.

Detailed Protocol: Analyzing BCR Repertoire Evolution Post-Vaccination

Software: IgPhyML, BEAST2 with BDSIR package, dynamice.

Workflow:

  • Longitudinal BCR Repertoire Sequencing: Isolate PBMCs from pre-vaccination (day 0) and multiple post-vaccination timepoints (e.g., day 7, 14, 28). Perform heavy-chain (IGH) repertoire sequencing (e.g., via 5'RACE or multiplex PCR).
  • Clonal Lineage Definition: Use Change-O or IgBlast to annotate sequences and group them into clonal lineages based on shared V/J genes and CDR3 similarity.
  • Build Lineage Trees: For each expanded clone, construct a phylogenetic tree of its somatic hypermutated variants.
  • Bayesian Evolutionary Analysis:
    • For Selection: Use IgPhyML to fit codon models and estimate dN/dS (ω) per branch or site.
    • For Population Dynamics: Use a structured birth-death model in BEAST2 (BDSIR) on lineage counts over time to jointly infer the birth (expansion) and death (contraction) rates of specific B-cell clades.
  • Infer Dynamics: Plot posterior distributions of λ and μ for antigen-responsive clones versus naive background.

The Scientist's Toolkit: Antibody Repertoire Development

Table 6: Essential Research Reagents & Tools

Item Function/Description
5' RACE-based BCR Seq Kit (e.g., SMARTer Human BCR) Captures full-length, unbiased BCR transcripts for repertoire analysis.
MiXCR / IgBlast Software for processing raw BCR sequencing reads, assigning V(D)J genes, and identifying clones.
Change-O & Alakazam Suite for advanced BCR repertoire analysis, lineage building, and diversity statistics.
IgPhyML Phylogenetic software designed to detect selection in BCR lineages using codon models.
ImmuneDB Database and analysis platform for managing and querying large adaptive immune receptor datasets.

G Blood Longitudinal Blood Draws BCRseq BCR Heavy-Chain Sequencing Blood->BCRseq Clones Clonal Lineage Definition BCRseq->Clones LineageTree Per-Clone Phylogenetic Trees Clones->LineageTree Analysis Bayesian Analysis LineageTree->Analysis Sel Selection (dN/dS) Analysis->Sel Dyn Population Dynamics (λ, μ) Analysis->Dyn

Title: BCR Repertoire Dynamics Analysis Pathway

Implementing Bayesian Birth-Death Models: A Step-by-Step Guide for Phylogenetic Analysis

Within Bayesian birth-death skyline modeling for diversity history research, time-stamped molecular sequences are the foundational data. These models estimate rates of speciation (birth), extinction (death), and sampling through time from phylogenetic trees with known node ages. The accuracy of these inferences is critically dependent on the quality and precision of the temporal (sampling date) and molecular data. This protocol details the requirements and preparation of such data from two key sources: rapidly evolving pathogens (e.g., viruses) and somatic cell populations (e.g., tumor biopsies).

Core Data Requirements & Specifications

High-fidelity input data must satisfy the criteria in Table 1.

Table 1: Core Data Requirements for Bayesian Birth-Death Analysis

Data Component Specification Purpose in Birth-Death Analysis
Molecular Sequence High-quality consensus sequence for each sample (FASTA format). Minimum coverage ≥100x for NGS data. Provides the phylogenetic signal for tree reconstruction.
Sampling Date Exact calendar date (YYYY-MM-DD) for each sequence. Critical precision. Anchors tips of the phylogenetic tree in time, enabling rate calibration.
Sequence Metadata Host ID, location, clinical stage (for tumors), subtype/clade. Enables covariate analysis (e.g., testing if birth rate varies with location).
Alignment Codon-aware for coding regions; gaps and ambiguities minimized. Ensures homology for accurate phylogenetic model likelihood calculation.
Data Completeness <5% ambiguous bases (N's) per sequence. Reduces phylogenetic uncertainty and computational artifacts.

Application Notes & Protocols

Protocol A: Preparing Time-Stamped Viral Sequences from NGS Data

Objective: Generate accurate consensus sequences with precise sampling dates from viral isolate sequencing.

Materials & Workflow:

  • Raw Read Processing:
    • Tool: FastQC, Trimmomatic.
    • Method: Assess read quality. Trim adapters and low-quality bases (Phred score <30).
  • Alignment & Consensus Calling:
    • Tool: BWA-MEM or Bowtie2 for alignment; SAMtools/BCFtools for variant calling.
    • Method: Map reads to a reference genome. Generate consensus sequence using a majority-rule threshold (e.g., base call supported by ≥75% of reads at coverage ≥100x). Call minor variants only if within-host diversity is of interest.
  • Temporal Annotation:
    • Method: Extract collection date from sample metadata. Convert to decimal date (e.g., 2023-04-15 → 2023.29) for phylogenetic software compatibility.
  • Multiple Sequence Alignment (MSA):
    • Tool: MAFFT or Clustal Omega.
    • Method: Align all consensus sequences. Visually inspect (e.g., with AliView) and trim to common coding regions.

viral_protocol raw_reads Raw NGS Reads (FASTQ) qc_trim Quality Control & Adapter Trimming raw_reads->qc_trim aligned Aligned Reads (BAM) qc_trim->aligned consensus Consensus Sequence (FASTA) aligned->consensus msa Multiple Sequence Alignment consensus->msa meta Sample Metadata date Decimal Date Annotation meta->date date->msa output Time-Stamped Aligned Sequences msa->output

Diagram Title: Viral Sequence Preparation Workflow

Protocol B: Preparing Time-Stamped Sequences from Serial Tumor Biopsies

Objective: Extract and prepare somatic variant profiles (e.g., from specific genes) from longitudinal tumor samples for phylogenetic birth-death analysis of clonal evolution.

Materials & Workflow:

  • Tissue Processing & NGS:
    • Method: Macro-dissection of FFPE or frozen tumor tissue. DNA extraction. Target enrichment (hybrid-capture or amplicon) for a defined gene panel (e.g., cancer driver genes).
  • Variant Calling & Filtering:
    • Tool: GATK Mutect2 (for paired tumor/normal) or VarScan2.
    • Method: Call somatic single nucleotide variants (SNVs) and small indels. Apply strict filters: minimum alternate allele read depth ≥20, variant allele frequency (VAF) ≥5%, and presence in ≥2 replicates if available.
  • Variant Alignment & Binary Encoding:
    • Method: Create a binary sequence for each sample where each character represents a genomic position in the panel. Use '1' for mutant, '0' for wild-type, '?' for missing data. This creates a "sequence" of somatic states.
  • Temporal Annotation:
    • Method: Assign each sample a time point relative to therapy initiation (e.g., Day 0 [baseline], Day 90 [progression]). This serves as the sampling date for birth-death models of clonal expansion/decay.

tumor_protocol biopsy Serial Tumor Biopsies ngs Targeted NGS (Gene Panel) biopsy->ngs variants Somatic Variant Calls (VCF) ngs->variants filter Strict Variant Filtering variants->filter encode Binary Sequence Encoding filter->encode output_b Time-Stamped Variant Sequences encode->output_b time_rel Annotate Relative Time Points time_rel->output_b

Diagram Title: Tumor Variant Sequence Preparation

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials for Time-Stamped Sequence Preparation

Item Function Example Product/Kit
High-Fidelity DNA Polymerase Accurate amplification of template material for sequencing libraries, minimizing PCR errors that confound phylogenetics. Q5 High-Fidelity DNA Polymerase (NEB).
Hybrid-Capture Target Enrichment Kit Isolates sequences of interest (e.g., viral genome, cancer gene panel) from complex genomic background. xGen Hybridization Capture Kit (IDT); SureSelectXT (Agilent).
Ultra-Low Input Library Prep Kit Constructs sequencing libraries from minute quantities of input DNA (critical for degraded FFPE tumors). SMARTer ThruPLEX Plasma-Seq (Takara Bio); KAPA HyperPrep.
Multiplexing Index Adapters Allows pooling of multiple samples in one sequencing run, ensuring consistent processing and reducing batch effects. IDT for Illumina UD Indexes; TruSeq CD Indexes.
Reference Genome Material Positive control for alignment and variant calling (e.g., well-characterized cell line DNA for tumors). Genome in a Bottle Reference Materials (NIST).
Data Integrity Software Tracks and maintains immutable links between sample identifier, raw data, metadata, and analysis versions. Labvantage LIMS; openBIS.

Data Validation & Pre-Analysis Checklist

Before initiating Bayesian birth-death analysis, validate prepared data:

  • Temporal Signal Test: Perform a regression of root-to-tip genetic distance against sampling date (TempEst). A significant positive correlation (p < 0.05) confirms sufficient temporal signal.
  • Recombinant Screening: Use tools like RDP5 to detect and remove recombinant sequences (viruses) which violate tree-like evolutionary assumptions.
  • Metadata Consistency: Ensure all sequences in the alignment have unambiguous, accurate sampling dates.

Application Notes and Protocols

Within the broader thesis on Bayesian birth-death analysis for reconstructing phylogenetic diversity history, selecting and applying the appropriate software toolkit is paramount. This guide provides practical protocols for three core platforms: BEAST2, RevBayes, and Taming, each offering distinct approaches to modeling speciation and extinction dynamics.

Toolkit Comparison and Quantitative Benchmarks

Table 1: Comparison of Software Toolkits for Bayesian Birth-Death Analysis

Feature BEAST2 RevBayes Taming
Core Architecture Modular, plugin-based (BEAST 2 Core) Integrated, script-driven interpreter Standalone GUI/command-line for specific models
Primary Strength Rich ecosystem for complex, integrative models (e.g., phylodynamics). User-friendly GUIs (BEAUti). Unparalleled flexibility for custom model specification. Built-in MCMC, HMC, and VB inference. Specialized, efficient, and user-friendly for large trees under the Fossilized Birth-Death (FBD) model.
Model Specification XML-based, often generated via BEAUti. Direct, declarative scripting in Rev language. Configuration file or GUI input.
Inference Methods MCMC (via BEAST Core). MCMC, Hamiltonian Monte Carlo (HMC), Variational Bayes. Analytic likelihood calculations, MCMC.
Best For Standardized, published models; combining birth-death with sequence evolution & dating. Novel model development, pedagogical understanding, and bespoke analyses. Large-scale analyses under the FBD model with stratigraphic range data.
Typical Run Time High (for complex integrative models) Medium to High (flexibility trades off with optimization) Low to Medium (optimized for its specific model)
Learning Curve Moderate (steep for custom XML) Steep (requires scripting) Gentle

Table 2: Example Performance Metrics on a Simulated Dataset (100 Taxa, 5000 MCMC steps)

Metric BEAST2 (BDSS) RevBayes (FBD) Taming (FBD)
Wall-clock Time ~45 min ~30 min ~10 min
ESS (Effective Sample Size) for Net Diversification 320 410 500
Mean Posterior Speciation Rate (λ) 0.22 (0.15-0.30) 0.21 (0.14-0.29) 0.22 (0.15-0.30)

Detailed Experimental Protocols

Protocol 1: Setting up a Fossilized Birth-Death (FBD) Analysis in BEAST2

  • Data Preparation: Prepare a NEXUS file with aligned sequence data (e.g., data.nex) and a fossil calibration file specifying fossil taxa and their stratigraphic age ranges.
  • Model Specification via BEAUti: a. Launch BEAUti. Import data.nex. b. Navigate to the "Tip Dates" tab. Set "Use tip dates" and specify fossil and extant tip ages (0 for extant). c. Navigate to the "Priors" tab. Select "Birth Death Skyline Serial" or "Fossilized Birth Death" model from the "Tree Prior" dropdown. d. Add appropriate priors for diversification (speciation, extinction), fossil sampling (psi), and clock models. e. Generate the XML file (e.g., fbd_analysis.xml).
  • Execution & Diagnostics: a. Run the analysis in BEAST2: beast fbd_analysis.xml. b. Check MCMC convergence using Tracer (ESS > 200). c. Annotate the maximum clade credibility tree using TreeAnnotator.

Protocol 2: Scripting a Custom Birth-Death Model with Sampled Ancestors in RevBayes

  • Rev Script Structure: Create a script file my_fbd_analysis.Rev.
  • Define Core Parameters:

  • Specify the FBD Tree Prior:

  • Run MCMC:

Protocol 3: Conducting a Large-Scale FBD Analysis with Taming

  • Data Preparation: Prepare two files: a. A Newick tree file (tree.tre) with all taxa (extant and fossil). b. A corresponding age data file (ages.txt) with the first and last appearance dates for each taxon.
  • Configure Analysis: Use the Taming GUI or create a configuration file (taming_config.txt):

  • Execution: Run from the command line: taming taming_config.txt.
  • Output: Analyze the output log files in Tracer. Tree files can be visualized in FigTree.

Visualized Workflows and Relationships

G Start Start: Input Data BEAST2 BEAST2 Workflow Start->BEAST2 RevBayes RevBayes Workflow Start->RevBayes Taming Taming Workflow Start->Taming SubBEAST2 BEAUti -> XML Configuration BEAST2->SubBEAST2 SubRev Rev Script Model Definition RevBayes->SubRev SubTaming Config File / GUI Setup Taming->SubTaming End End: Posterior Estimates DataSeq Sequence Alignment (nex, phylip) DataSeq->BEAST2 DataSeq->RevBayes DataTree Tree + Fossil Ages (newick, txt) DataTree->Taming DataFossil Fossil Occurrences (txt) DataFossil->BEAST2 DataFossil->RevBayes Infer Run Bayesian Inference (MCMC) SubBEAST2->Infer SubRev->Infer SubTaming->Infer Check Diagnostics & Summarize Output Infer->Check Check->End

Title: Comparative Workflow of Three Birth-Death Analysis Toolkits

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 3: Essential Digital Research Reagents for Bayesian Birth-Death Analysis

Item (Software/Script) Function Primary Use Case
BEAUti (BEAST2) Graphical model specification generator. Produces the XML configuration file required to run BEAST2. Setting up standardized, complex integrative models without manual XML coding.
Tracer MCMC output analysis and diagnostics. Assesses convergence (ESS, trace plots) and summarizes parameter posterior distributions. Mandatory post-analysis for all toolkits to validate MCMC performance and interpret results.
FigTree / DensiTree Phylogenetic tree visualization. Renders maximum clade credibility trees (FigTree) or posterior tree distributions (DensiTree). Visualizing the estimated time-scaled phylogeny with node bars representing uncertainty.
Rev Scripts Custom, reproducible model definitions. Encapsulates the entire statistical model, priors, and inference setup in RevBayes. Developing novel models, teaching statistical concepts, and ensuring full analysis transparency.
Taming Configuration File Simple input for the Taming software. Specifies tree file, age data, and MCMC settings in a straightforward format. Efficiently configuring large-scale FBD analyses without complex scripting or GUI navigation.
TreeAnnotator (BEAST2) Summarizes the posterior sample of trees. Produces a single maximum clade credibility tree with summarized node heights/branch lengths. Generating the final, representative tree for publication from a BEAST2 posterior.

Application Notes

Within the framework of a thesis on Bayesian birth-death analysis for diversity history research, precise model specification is paramount. This process, typically executed in BEAST 2 (Bayesian Evolutionary Analysis Sampling Trees) via XML or specialized packages like BDSKY (Birth-Death Skyline), defines the generative process for phylogenetic trees and sequence evolution. Accurate configuration directly impacts inferences on speciation (birth), extinction (death), and sampling rates, which are critical for reconstructing the historical dynamics of viral epidemics, cancer cell lineages, or species diversification.

The core model is the birth-death sampling process, which generates the tree topology and node times. This is coupled with a molecular clock model that describes the rate of sequence evolution along branches and a substitution model for the sequence data itself. For researchers and drug development professionals, these models can test hypotheses about how therapeutic interventions or environmental changes alter pathogen population dynamics.

Table 1: Core Model Components in Bayesian Birth-Death Analysis

Component XML Element (BEAST 2) Key Parameters Scientific Purpose
Birth-Death Process BirthDeathSkylineModel R0 (reproductive number), becoming-uninfectious rate, sampling proportion. Estimates time-varying effective reproduction number and epidemic trajectory.
Sampling Process SerialSamplingModel, SAmpledAncestors Sampling proportion, sampling rate. Accounts for heterogeneous or serially-timed sample collection.
Molecular Clock StrictClockModel, RelaxedClockLogNormal Clock rate (mean, sigma). Calibrates evolutionary timeline; relaxed models account for rate variation.
Substitution Model HKY, GTR Kappa, base frequencies, substitution rates. Models the process of nucleotide/amino acid change.
Tree Prior BirthDeathModel Diversification rate, turnover, sampling probability. Provides prior probability distribution for tree topology and node ages.

Table 2: Representative Parameter Estimates from a Viral Phylogenetic Study

Parameter Prior Distribution Posterior Mean (95% HPD) Interpretation
R0 at origin LogNormal(M=1, S=1.5) 2.1 (1.4 - 3.0) Initial reproductive number.
Becoming-uninfectious rate (δ) Gamma(α=3, β=1) 0.5 yr⁻¹ (0.3 - 0.8) Rate of lineage loss (death+recovery).
Sampling proportion (ρ) Beta(α=1, β=1) 0.05 (0.01 - 0.12) Fraction of cases sequenced.
Clock rate (mean) LogNormal(M=-5, S=1) 8e⁻⁴ subs/site/yr (6e⁻⁴ - 1e⁻³) Average evolutionary rate.

Experimental Protocols

Protocol 1: Specifying a Birth-Death Skyline Model in BEAST 2 XML

Objective: Configure a time-varying birth-death model for analyzing pandemic virus phylogeny.

  • Define the Tree Prior: In the XML, inside the run block, specify the BirthDeathSkylineModel as the tree prior. Link it to the Tree element.
  • Parameterize Epidemic Dynamics: Set the reproductiveNumber parameter. This can be a single value or a piecewise constant function (Skyline) with dimension N (e.g., dimension="5") to allow N-1 change points.
  • Set Removal Rate: Define the becomeUninfectiousRate, typically as a single estimated value.
  • Configure Sampling: Specify the samplingProportion as a fixed value (if known) or an estimated parameter. For serial sampling, ensure type="serial" is set.
  • Set Origin Time: The origin parameter (time of the most recent common ancestor) requires a prior, often a gamma or log-normal distribution based on external data.
  • Link to Data: Ensure the tree attribute points to the @tree element defining the initial phylogeny.

Protocol 2: Calibrating Evolutionary Timeline with Clock Models

Objective: Apply a relaxed molecular clock to account for rate heterogeneity among branches.

  • Select Clock Model: Choose a RelaxedClockLogNormal model for uncorrelated rate variation among branches.
  • Set Mean Clock Rate: Specify the clock.rate parameter. This can be assigned a prior distribution (e.g., a log-normal with mean informed by literature).
  • Parameterize Rate Variation: The S (or ucldStdev) parameter controls the degree of among-branch rate variation. Assign a prior, such as Exponential(mean=0.5).
  • Integrate with Substitution Model: The clock model is nested within the site model, which contains the substitution model (e.g., HKY with kappa and frequencies parameters).

Protocol 3: Running a Bayesian MCMC Analysis

Objective: Execute the phylogenetic inference to obtain posterior distributions of model parameters.

  • Assemble XML: Combine the specified birth-death, clock, substitution, and data (alignment) elements into a complete BEAST 2 XML file. Use BEAUti (GUI) or beastlie (script) to aid generation.
  • Configure MCMC: Set chain length (chainLength) to an appropriate value (e.g., 50-100 million steps). Define logging frequencies for parameters (logEvery) and trees (treeLogEvery).
  • Execute Run: Run BEAST 2 from the command line: beast -threads 4 model_specification.xml.
  • Diagnose Convergence: Analyze output log files in Tracer to ensure ESS (Effective Sample Size) values >200 for all key parameters.
  • Summarize Trees: Use TreeAnnotator to generate a maximum clade credibility tree from the posterior tree set, discarding an appropriate burn-in percentage.

Visualizations

G Start Start: Alignment & Time Data BD Birth-Death- Sampling Model Start->BD Clock Molecular Clock Model Start->Clock TreePrior Tree Prior (Generated Tree) BD->TreePrior Clock->TreePrior Subst Substitution Model Subst->TreePrior MCMC MCMC Sampling TreePrior->MCMC Posterior Posterior Distributions: - R0(t), Rates, Tree MCMC->Posterior

Bayesian phylogenetic model specification workflow.

Birth-death-sampling model state transitions.

The Scientist's Toolkit

Table 3: Key Research Reagent Solutions for Bayesian Birth-Death Analysis

Item Function in Analysis Example/Format
BEAST 2 Core Software package providing the statistical framework for Bayesian phylogenetic analysis. Executable .jar file.
BDSKY Package BEAST 2 add-on implementing the Birth-Death Skyline model for serially sampled data. BEAST 2 package.
Sequence Alignment Input molecular data (FASTA format) for the taxa of interest. .fasta or .nexus file.
Tracer Graphical tool for analyzing MCMC output, assessing convergence (ESS), and summarizing posteriors. Application.
TreeAnnotator Summarizes posterior tree distributions into a single target tree (Maximum Clade Credibility). BEAST 2 utility.
FigTree / IcyTree Visualizes and annotates the resulting phylogenetic trees. Application.
BEAUti Graphical interface to generate BEAST XML configuration files from alignments and trait data. BEAST 2 utility.
Clock Rate Prior Informative prior distribution for the molecular clock rate, derived from literature or calibration points. e.g., LogNormal(M=-5, S=0.8).

Within a Bayesian birth-death analysis framework for reconstructing species diversification histories, Markov Chain Monte Carlo (MCMC) sampling is the computational engine for approximating posterior distributions of parameters like speciation (λ) and extinction (μ) rates. The reliability of these inferences hinges entirely on proper MCMC configuration and rigorous assessment of chain convergence and sampling efficiency.

Core MCMC Settings for Birth-Death Analyses

The following table summarizes critical MCMC settings, their typical configurations, and their roles in ensuring robust sampling for phylogenetic birth-death models.

Table 1: Standard MCMC Settings for Bayesian Birth-Death Analyses

Parameter/Setting Typical Value/Range Function & Rationale
Number of Chains 2-4 independent chains Enables assessment of convergence via inter-chain statistics (e.g., R-hat).
Chain Length (Generations) 10⁷ - 10⁸ (data-dependent) Must be sufficiently long for chains to explore the posterior space thoroughly.
Burn-in Period 10-50% of total generations Initial samples discarded before chains have stabilized at the target distribution.
Sampling Frequency (Thinning) Every 10³ - 10⁴ steps Reduces autocorrelation in saved samples and manages file size. Use with caution.
Proposal Mechanism Adaptive Metropolis, Sliding Window, Scale Operators Algorithms for proposing new parameter states. Tuning acceptance rates is crucial.
Target Acceptance Rate 20-40% (for continuous parameters) Optimizes chain mixing; too high/low indicates poorly tuned proposal distributions.

Protocol: Configuring and Executing an MCMC Run for BAMM/RevBayes

This protocol outlines a standard workflow for running a diversification rate analysis using BAMM or RevBayes, common software in the field.

A. Pre-Run Configuration

  • Model Specification: Define the full birth-death model (e.g., time-dependent, rate-heterogeneous). Specify priors for all parameters (λ, μ, rate shift number).
  • MCMC Initialization: Set the number of independent chains (nruns=2 or nchains=4). Initialize chains from random points in parameter space.
  • Run Parameters: Set the total number of generations (ngen=50,000,000). Define the sampling interval (sampleFreq=10000) and burn-in (burnin=10,000,000).
  • Proposal Tuning: Perform a short pilot run (e.g., 1-2 million generations). Analyze the acceptance rates of proposal operators and adjust their tuning parameters (e.g., lambda) to achieve acceptance rates within the 20-40% window.

B. Execution & Monitoring

  • Run MCMC: Execute the analysis on a high-performance computing cluster. Redirect output log files for each chain separately.
  • Real-time Monitoring: Use tools like Tracer to monitor trace plots and ESS values during the run to identify obvious failures early.

C. Post-Run Diagnostics

  • Log File Inspection: Confirm run completed without errors. Verify the specified number of samples were written.
  • Convergence Assessment: Proceed to the diagnostic steps outlined in Section 4.

Diagram 1: MCMC Setup and Execution Workflow

workflow Start Define Birth-Death Model & Priors Config Set MCMC Parameters (Chains, Length, Sampling) Start->Config Pilot Run Short Pilot MCMC Config->Pilot Tune Tune Proposal Operators Pilot->Tune Pilot->Tune Check Acceptance Rates FullRun Execute Full MCMC Run Tune->FullRun Monitor Real-Time Monitoring (Trace Plots, ESS) FullRun->Monitor Monitor->FullRun If poor, stop and adjust Diagnose Post-Run Convergence Diagnostics Monitor->Diagnose

Chain Convergence Diagnostics

Convergence indicates that multiple MCMC chains have sampled from the same target posterior distribution. It is assessed using multiple criteria.

Table 2: Key Convergence Diagnostics and Their Interpretation

Diagnostic Target Value Calculation & Interpretation
Potential Scale Reduction Factor (R-hat / Gelman-Rubin) ≤ 1.01 (≤ 1.05 acceptable) Ratio of between-chain to within-chain variance. Values >>1 indicate non-convergence.
Average Standard Deviation of Splits (ASDSF) < 0.01 Measures topological convergence in phylogenetic analyses by comparing tree samples.
Trace Plots (Visual) Stationary, "fuzzy caterpillar" Visual inspection of parameter values across generations. Should show stable mean and variance.
Effective Sample Size (ESS) > 200 (per parameter) See Section 5. Quantifies independent samples. Low ESS indicates high autocorrelation.

Protocol: Performing Convergence Diagnostics with R and coda

  • Load Data: Import the MCMC log files (e.g., chain1.log, chain2.log) into R using the coda or rstan packages.

  • Generate Trace and Density Plots: Visually assess mixing and stationarity.

  • Calculate R-hat: Compute the Gelman-Rubin diagnostic for all key parameters.

  • Calculate ESS (Initial): Obtain an initial estimate of sampling efficiency (see next section).

Diagram 2: Convergence Diagnostic Decision Pathway

decision A R-hat ≤ 1.01 for all parameters? B ESS > 200 for all parameters? A->B Yes E RUN REJECTED Investigate Cause A->E No C Trace Plots show stationary mixing? B->C Yes B->E No D CONVERGENCE ACCEPTED C->D Yes C->E No Start Start Start->A

Effective Sample Size (ESS)

ESS estimates the number of independent samples equivalent to the autocorrelated MCMC samples. It is the most critical metric for determining whether posterior estimates (mean, HPD) are reliable.

Table 3: ESS Interpretation and Troubleshooting

ESS Value Interpretation Recommended Action
ESS > 200 Sufficient for reliable mean estimates. Proceed with analysis.
ESS > 1000 Good for reliable estimates of 95% HPD intervals. Ideal scenario.
ESS < 200 Insufficient. Parameter estimates are unreliable. Must increase effective sampling.

Protocol: Diagnosing and Remedying Low ESS

  • Diagnose Cause: Calculate autocorrelation times and plots.

  • Apply Remedies:
    • Increase Run Length: Extend MCMC by a factor of (Target ESS / Current ESS).
    • Improve Proposal Mechanisms: Re-tune proposal distributions based on acceptance rates from the initial run.
    • Re-parameterize Model: Simplify the model or use alternative parameterizations (e.g., centered vs. non-centered) to reduce posterior correlations.
    • Use Alternative Algorithms: Consider integrating Hamiltonian Monte Carlo (HMC) or No-U-Turn Sampler (NUTS) via software like RevBayes or Stan for complex models.

The Scientist's Toolkit: Essential Research Reagents & Software

Table 4: Key Software and Analytical Tools for MCMC in Diversification Studies

Item Function Example/Provider
MCMC Sampling Engine Core software for performing Bayesian inference. RevBayes, BAMM, MrBayes, STAN (via PhyloStan)
Diagnostic & Visualization Assessing convergence, mixing, and ESS. Tracer, R packages coda, rstan, boa
High-Performance Computing (HPC) Infrastructure for running long, multi-chain analyses. University clusters, NSF XSEDE, Cloud computing (AWS, GCP)
Phylogenetic Data Time-calibrated trees for analysis. TreeBASE, Open Tree of Life, bespoke Bayesian dating analyses (BEAST2)
Scripting Environment Automating analyses, processing logs, generating reports. R, Python, bash shell scripts

Application Notes

Bayesian birth-death (BD) skyline models provide a powerful phylogenetic framework to reconstruct the dynamic spread and diversification of pandemic viruses like SARS-CoV-2. By analyzing time-scaled viral phylogenies, these models estimate time-varying effective reproduction numbers (Re) and become non-informative (sampling proportions) through time. This application is critical for quantifying the impact of public health interventions, host adaptation, and immune evasion on viral lineage birth (transmission) and death (recovery/sampling) rates.

Key Inferences:

  • Temporal Shifts in Transmission: BD models can identify periods of expansion (high Re) and decline (low Re) for specific variants, correlating with real-world events.
  • Comparative Lineage Dynamics: The relative fitness of Variants of Concern (VOCs) can be compared by contrasting their estimated birth rates.
  • Sampling Bias Assessment: The model jointly estimates the rate of sequence sampling through time, which is crucial for correcting surveillance biases in diversity reconstruction.

Table 1: Estimated Evolutionary and Epidemiological Parameters for Select SARS-CoV-2 Variants (Illustrative)

Variant (Pango Lineage) Approx. Emergence Date Mean Evolutionary Rate (subs/site/year) Estimated Peak Re (Birth Rate) Period of Dominance Key Spike Mutations
Alpha (B.1.1.7) Sep-2020 ~1.1 x 10^-3 1.5 - 1.8 Dec-2020 to May-2021 N501Y, Δ69-70, P681H
Delta (B.1.617.2) Oct-2020 ~1.0 x 10^-3 1.8 - 2.2 Jun-2021 to Dec-2021 L452R, T478K, P681R
Omicron BA.1 (B.1.1.529.1) Nov-2021 ~1.4 x 10^-3 2.0 - 2.5 Dec-2021 to Mar-2022 G339D, S371L, S477N, Q498R
Omicron BA.2 (B.1.1.529.2) Nov-2021 ~1.4 x 10^-3 1.6 - 2.0 Feb-2022 to May-2022 T376A, D405N, R408S

Table 2: Core Input Data for Bayesian Birth-Death Analysis

Data Type Description Source Example Purpose in Analysis
Viral Genome Sequences Time-stamped, high-coverage whole genomes. GISAID, NCBI Virus Build time-resolved phylogeny.
Sequence Metadata Collection date, location, host. GISAID Calibrate molecular clock, stratify analysis.
Epidemiological Data Case counts, vaccination rates. WHO, Our World in Data Contextual validation of model estimates.

Experimental Protocols

Protocol 1: Time-Scaled Phylogenetic Reconstruction for Birth-Death Analysis

Objective: To infer a rooted, time-scaled phylogenetic tree from SARS-CoV-2 sequence data for downstream BD modeling.

Materials: High-performance computing cluster, BEAST 2.x software suite, Tracer v1.7+, IcyTree v1.0.0+.

Procedure:

  • Data Curation & Alignment: Download a globally representative subset of SARS-CoV-2 genomes with collection dates (e.g., ~500 sequences per major variant). Perform multiple sequence alignment using MAFFT or Nextclade.
  • Substitution Model Selection: Use ModelTest-NG or bModelTest within BEAST to determine the best-fit nucleotide substitution model (e.g., GTR+G+I).
  • Clock Model and Tree Prior Setup: Configure the analysis with:
    • A strict or relaxed (uncorrelated lognormal) molecular clock model.
    • A non-parametric Bayesian Skyline tree prior as an initial flexible model.
  • MCMC Run: Execute two independent Markov Chain Monte Carlo (MCMC) runs for at least 100 million generations, sampling every 10,000 steps.
  • Log File Diagnostics: Use Tracer to assess convergence (ESS > 200 for all parameters), appropriate burn-in, and stationarity.
  • Maximum Clade Credibility (MCC) Tree Generation: Use TreeAnnotator to combine post-burn-in trees from both runs and generate a single time-scaled MCC tree.

Protocol 2: Bayesian Birth-Death Skyline Model Analysis

Objective: To estimate time-varying effective reproduction numbers (Re) and sampling proportions from the time-scaled phylogeny.

Materials: BEAST 2.x with BDMM package, R with rBEAST and ggplot2 packages.

Procedure:

  • Model Specification: Load the MCC tree into a new BEAST analysis. Select the "Birth-Death Skyline Serial" model.
  • Parameterization:
    • Set the number of time intervals for the skyline (e.g., 10 intervals over the pandemic timeline).
    • Define the origin prior (e.g., based on the estimated tMRCA of the pandemic).
    • Specify an informed prior for the sampling proportion based on reported case ascertainment rates.
  • MCMC Run: Execute MCMC for 50 million generations, sampling every 5,000 steps.
  • Parameter Estimation: In Tracer, examine the posterior distributions for R0 (effective reproduction number) and samplingProportion for each time interval.
  • Visualization: Use the rBEAST library in R to plot the median and 95% HPD (Highest Posterior Density) intervals of R0 through time, overlaying key variant emergence dates.

Visualizations

bd_workflow seq Time-Stamped Viral Sequences align Multiple Sequence Alignment seq->align tree_prior Tree & Clock Model Selection align->tree_prior mcmc1 MCMC Phylogenetic Inference tree_prior->mcmc1 mcc Time-Scaled MCC Tree mcmc1->mcc bd_model Birth-Death Skyline Model Setup mcc->bd_model mcmc2 MCMC BD Parameter Estimation bd_model->mcmc2 re_plot Time-Varying R(t) Plot mcmc2->re_plot

Title: Phylogenetic & Birth-Death Analysis Workflow

bd_logic Transmission Transmission BirthRate Birth Rate (λ) ≈ Re * δ Transmission->BirthRate Lineage Birth Recovery Recovery DeathRate Death Rate (μ + ψ) Recovery->DeathRate Lineage Death (μ) Sampling Sampling Sampling->DeathRate Lineage Death (ψ) SampledTree Sampled Phylogenetic Tree BirthRate->SampledTree DeathRate->SampledTree

Title: Birth-Death Process & Phylogeny Relationship

The Scientist's Toolkit

Table 3: Key Research Reagent Solutions for Viral Diversification Studies

Item Function/Description Example Product/Resource
Viral Transport Medium (VTM) Preserves viral RNA integrity during clinical sample transport. Copan UTM, CDC-recommended VTM formula.
Whole Genome Sequencing Kit For amplification and library prep of viral genomes from low-titer samples. ARTIC Network v4.1 primer pools & Illumina COVIDSeq Test.
Metagenomic RNA Library Prep Kit Enables unbiased sequencing for variant detection in complex samples. Illumina Respiratory Virus Oligo Panel, QIAseq DIRECT SARS-CoV-2.
Phylogenetic Software Suite Performs alignment, model testing, and Bayesian phylogenetic inference. BEAST 2.7, IQ-TREE 2.2.0, Nextclade CLI.
High-Performance Computing (HPC) Resource Essential for computationally intensive MCMC analyses and large dataset handling. Local HPC cluster, Cloud computing (AWS, GCP).
Curated Sequence Database Provides essential, quality-controlled sequence data with metadata. GISAID EpiCoV database, NCBI Virus SARS-CoV-2 Resources.

Overcoming Challenges in Bayesian Birth-Death Analysis: Prior Sensitivity, Model Misspecification, and Computational Hurdles

This application note is a component of a broader thesis investigating Bayesian birth-death models for elucidating diversity histories in evolutionary biology, with direct applications to pathogen evolution and drug target discovery. A central tenet of Bayesian inference is the formal incorporation of prior knowledge through the prior probability distribution. The choice between informative and vague (diffuse) priors for diversification rate parameters (speciation, λ, and extinction, μ) is non-trivial and critically shapes posterior estimates. This document outlines the methodological considerations, provides experimental protocols for sensitivity analysis, and presents current data on their impacts.

Core Concepts and Quantitative Comparisons

Table 1: Characteristics of Informative vs. Vague Priors for Diversification Rates

Prior Type Typical Distribution Parameterization Example Justification Primary Risk
Informative Prior Lognormal, Gamma Lognormal(meanlog=0.1, sdlog=0.5) for λ. Based on previous empirical studies (e.g., fossil-calibrated rates for clade). Prior-biased posteriors if prior knowledge is incorrect.
Vague/Diffuse Prior Exponential, Uniform Exponential(rate=0.1) or Uniform(0, 100). Represents minimal knowledge; lets data dominate. Inefficient sampling, overly broad credible intervals.
Empirical Hyperprior Hyperlognormal Mean from meta-analysis of rates, with hyperprior on variance. Hierarchical borrowing of strength across studies. Computational complexity.

Table 2: Impact of Prior Choice on Posterior Estimates (Synthetic Data Example)

Simulated True Value Prior Type Posterior Mean (95% HPD) ESS Posterior SD
λ = 0.2, μ = 0.1 Vague: Exp(1.0) λ: 0.22 (0.05, 0.45), μ: 0.12 (0.01, 0.30) 120 0.10
λ = 0.2, μ = 0.1 Informative: Lognorm(ln(0.2), 0.3) λ: 0.21 (0.15, 0.28), μ: 0.11 (0.05, 0.18) 450 0.03
λ = 0.2, μ = 0.1 Misinformative: Lognorm(ln(0.5), 0.3) λ: 0.38 (0.30, 0.47), μ: 0.11 (0.05, 0.18) 400 0.04

Experimental Protocols

Protocol 3.1: Prior Sensitivity Analysis for Birth-Death Models

Objective: To quantify the sensitivity of posterior diversification rate estimates to different prior specifications. Materials: Phylogenetic tree (nexus format), Bayesian inference software (e.g., RevBayes, BEAST2, PyRate). Procedure:

  • Data Preparation: Import and condition the ultrametric timetree. Check for appropriate branch length units (e.g., millions of years).
  • Model Definition: Specify a time-homogeneous birth-death process as the tree prior. Fix sampling fraction (ρ) based on known taxonomy.
  • Prior Specification Sets:
    • Set A (Vague): Assign an Exponential(1.0) prior to both speciation (λ) and extinction (μ) rates.
    • Set B (Informative): Assign Lognormal priors. For λ, set meanlog=log(empiricalmeanfrom_literature), sdlog=0.5. Repeat for μ.
    • Set C (Misinformative): Assign Lognormal priors with meanlog deliberately offset from plausible values (e.g., double the empirical mean).
  • MCMC Execution: Run ≥ 3 independent MCMC chains for 50,000 generations each, sampling every 50. Use effective sample size (ESS) > 200 as convergence criterion.
  • Posterior Analysis: Compare marginal posterior distributions of λ and μ across Prior Sets A, B, and C using overlapping credible interval plots and Kullback-Leibler divergence. Deliverable: A sensitivity plot and a table comparing posterior summaries (Table 2 format).

Protocol 3.2: Calibrating Informative Priors from Meta-Analysis

Objective: To construct empirically justified informative priors from published diversification rate studies. Materials: Database of published rates (e.g., from literature search or compendium like www.timetree.org). Procedure:

  • Literature Search: Conduct a systematic search for empirical diversification rate estimates for a relevant taxonomic group (e.g., RNA viruses, mammals). Use keywords: "diversification rate," "birth-death model," "[clade name]," "speciation rate estimate."
  • Data Extraction: Extract point estimates and measures of uncertainty (SD, HPD) for λ and μ. Record the tree prior and calibration method used in each source.
  • Distribution Fitting: Log-transform the set of extracted λ (or μ) rates. Calculate the mean (μlog) and standard deviation (σlog) of these log-transformed values.
  • Prior Parameterization: Define the informative prior as Lognormal(meanlog = μlog, sdlog = σlog + 0.2), where the added 0.2 accounts for between-study heterogeneity.
  • Validation: Apply the derived prior in a cross-validation test, withholding a subset of studies, to check for calibration.

Visualization: Analytical Workflow

G Start Input: Time-Calibrated Phylogeny P2 Protocol 3.1: Define Prior Sets (A: Vague, B: Informative, C: Misinformative) Start->P2 P1 Protocol 3.2: Calibrate Informative Prior (Meta-Analysis) P1->P2 Provides Parameters P3 Run MCMC under Birth-Death Model P2->P3 P4 Assess Convergence (ESS > 200, PSFR ~1.0) P3->P4 P4->P3 Not Converged P5 Extract Posterior Distributions for λ, μ P4->P5 Converged P6 Comparative Analysis: - Credible Intervals - KL Divergence - Sensitivity Plot P5->P6 End Output: Decision on Robust Prior Choice P6->End

Workflow for Prior Sensitivity Analysis in Diversification Studies

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials and Software for Bayesian Diversification Analysis

Item Name / Solution Category Function / Application Example / Note
RevBayes v1.2.1 Software Modular platform for Bayesian phylogenetic analysis. Implements a wide range of birth-death models. Allows custom prior specification; essential for Protocol 3.1.
BEAST2 + BDMM Package Software Bayesian evolutionary analysis; BDMM adds structured birth-death models. Useful for host-associated pathogen diversification.
TreeAnnotator Software Summarizes posterior tree distribution into a single maximum clade credibility tree. Used post-MCMC to generate a representative tree.
Tracer v1.7.2 Software Diagnoses MCMC convergence and summarizes parameter posterior distributions. Calculates ESS, generates marginal density plots (Protocol 3.1, Step 5).
PyRate Software Bayesian analysis of fossil data; estimates speciation/extinction rates with time-variable models. Alternative for paleontological data; can inform prior calibration.
TimeTree Database Online Resource Public knowledge-base for species divergence times. Source for secondary calibration points and comparative rate data (Protocol 3.2).
Jupyter Notebook + R Computing Environment Reproducible environment for scripting analyses, parsing logs, and creating visualizations. Integrates steps from tree processing to final plotting.
High-Performance Computing (HPC) Cluster Infrastructure Runs computationally intensive MCMC analyses for large datasets or complex models. Necessary for analyses with genome-scale data or hundreds of taxa.

Diagnosing and Resolving Poor MCMC Mixing and Non-Convergence in Complex Models

1. Introduction within a Bayesian Birth-Death Analysis Thesis

Within the broader thesis investigating diversity history through Bayesian birth-death models, achieving reliable posterior inference is paramount. Complex models, such as the Fossilized Birth-Death (FBD) process with episodic rate shifts, often suffer from poor Markov Chain Monte Carlo (MCMC) mixing and non-convergence. This application note provides protocols to diagnose these issues and implement targeted solutions, ensuring the robustness of conclusions drawn about speciation, extinction, and sampling rates through deep time.

2. Diagnostic Table: Key Metrics and Their Interpretation

Table 1: Quantitative Metrics for Assessing MCMC Performance

Metric Calculation/Indicator Target Value Interpretation of Poor Values
Effective Sample Size (ESS) Posterior samples independent of autocorrelation. >200 per parameter (minimum). ESS < 200 indicates high autocorrelation, insufficient independent samples.
Gelman-Rubin Diagnostic (R̂) Variance between chains vs. within chains. ≤ 1.01 (strict), ≤ 1.05 (lenient). R̂ >> 1.05 indicates chains have not converged to the same distribution.
Trace Plot Visuality Visual inspection of parameter value vs. iteration. Stable fluctuation around a constant mean. Trends, sharp shifts, or lack of movement ("sticky" chains) indicate issues.
Autocorrelation Time Number of steps to produce an independent sample. As low as possible; high ESS. High autocorrelation indicates inefficient exploration of parameter space.
Monte Carlo Standard Error (MCSE) Uncertainty in the posterior mean estimate. Small relative to posterior standard deviation. Large MCSE suggests the mean estimate is imprecise.

3. Experimental Protocols for Diagnosis and Resolution

Protocol 3.1: Comprehensive MCMC Diagnostics Workflow

  • Run Multiple Chains: Initiate at least 4 MCMC chains from dispersed starting points (e.g., from random draws from the prior distribution).
  • Configure Samplers: For complex birth-death models, ensure adaptive operators are used (e.g., in BEAST2, AdaptiveOperatorSampler).
  • Run Length Determination: Conduct pilot runs. Use the Tracer software to assess ESS. If ESS < 200 for key parameters (e.g., diversification rates), extend run length by a factor of 10 or more.
  • Calculate Diagnostics: Compute R̂ and ESS for all numerical parameters using tools like Tracer or arViz in Python/R.
  • Visual Inspection: Generate trace plots, density plots, and autocorrelation plots for all parameters. Identify problematic parameters.

Protocol 3.2: Resolving Poor Mixing via Reparameterization

Objective: Improve sampler efficiency by transforming parameters to a space closer to multivariate normality.

  • Identify Correlated Parameters: Review correlation matrices or pairwise scatter plots from pilot runs (e.g., LogCombiner and Tracer output).
  • Apply Transformations:
    • For strictly positive parameters (e.g., birth rate λ, death rate μ), use a log-transform. Implement by placing a log-normal or exponential prior, or sampling on a log scale.
    • For parameters on the probability simplex (e.g., relative sampling rates across epochs), use a logit or generalized logit transform.
  • Re-run MCMC: Re-configure the model to sample the transformed parameter. Apply the inverse transform for interpretation.
  • Re-assess Diagnostics: Compare ESS and trace plots pre- and post-reparameterization to quantify improvement.

Protocol 3.3: Protocol for Path and Stepping-Out Sampling

Objective: Efficiently sample from complex, correlated posteriors common in time-varying birth-death models.

  • Path Sampling (Thermodynamic Integration):
    • Define a path from the prior (β=0) to the posterior (β=1) via a power posterior: p(θ|β) ∝ p(D|θ)^β p(θ).
    • Run MCMC for a series of β values (e.g., 0.0, 0.1,..., 1.0).
    • Calculate the marginal log-likelihood (model evidence) by integrating over β. Use this for model comparison or to check for implementation errors.
  • Stepping-Out Slice Sampler:
    • For parameters with difficult conditional distributions, replace standard Metropolis-Hastings with a slice sampler.
    • Algorithm: Given current point x0, draw a horizontal slice y ~ Uniform(0, f(x0)). Define an interval (L, R) around x0 by stepping out until f(L) < y and f(R) < y. Sample a new point x1 uniformly from this interval.
    • This adapts to scale and is effective for multi-modal distributions.

4. Visualization: Diagnostic and Remediation Workflow

workflow Start Initial MCMC Run (Complex Birth-Death Model) Diagnose Diagnostic Assessment Start->Diagnose ESS_OK ESS > 200 & R̂ ≤ 1.01? Diagnose->ESS_OK Conclude Reliable Posterior Proceed to Thesis Analysis ESS_OK->Conclude Yes ParamSpace Poor Mixing in Parameter Space ESS_OK->ParamSpace No DataLikelihood Poor Likelihood Evaluation ESS_OK->DataLikelihood No Reparam Remedy A: Reparameterization (e.g., log-transform) ParamSpace->Reparam AdvancedSamplers Remedy B: Advanced Samplers (Path/Slice/HMC) DataLikelihood->AdvancedSamplers ModelCheck Remedy C: Model/Prior Review & Data Partitioning DataLikelihood->ModelCheck Reassess Re-run MCMC & Re-assess Diagnostics Reparam->Reassess AdvancedSamplers->Reassess ModelCheck->Reassess Reassess->ESS_OK

Diagram 1: MCMC Diagnosis and Remediation Decision Workflow

5. The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Software and Analytical Tools for MCMC in Phylogenetics

Item (Software/Package) Primary Function Application in Birth-Death Analysis
BEAST2 / MrBayes Bayesian Evolutionary Analysis Platform. Core software for implementing MCMC on Fossilized Birth-Death and related models.
Tracer MCMC Diagnostic Visualization. Calculates ESS, R̂, examines trace plots, and posterior densities for all parameters.
TreeAnnotator Summarizes Posterior Tree Samples. Generates maximum clade credibility trees from posterior, integrating divergence times.
RevBayes Flexible Probabilistic Programming. Allows custom specification of complex, hierarchical birth-death models for research.
CODA / arViz R/Python Diagnostic Packages. Programmatic calculation of convergence diagnostics and custom visualizations.
PathSampler (BEAST2) Thermodynamic Integration. Estimates marginal likelihood for model comparison of different diversification scenarios.
DensiTree Visualization of Tree Distributions. Assesses convergence and uncertainty in posterior tree topology and node heights.
LogCombiner MCMC Log File Manipulation. Merges, subsamples, and reparametrizes log files from multiple runs for analysis.

Within the broader thesis on Bayesian birth-death analysis for diversity history research, a fundamental challenge is the reconciliation of observed phylogenetic data with the true, unknown evolutionary history. This observed data is almost invariably an incomplete sample of the lineages that have existed through time. This article details application notes and protocols for modeling and correcting these sampling biases, a critical step in producing robust estimates of speciation, extinction, and sampling rates from molecular phylogenies.

Foundational Concepts & Quantitative Frameworks

Sampling bias in phylogenetic birth-death models is formally incorporated via a sampling probability (ρ for extant species) and/or a Poisson sampling rate (ψ for fossils). The table below summarizes key parameters and data types used in contemporary models.

Table 1: Core Parameters and Data Types for Sampling-Bias Aware Birth-Death Models

Parameter/Symbol Typical Notation Description Data Type/Input Required
Speciation Rate λ (lambda) Rate at which lineages split. Estimated from phylogeny.
Extinction Rate μ (mu) Rate at which lineages go extinct. Estimated from phylogeny.
Extant Sampling Probability ρ (rho) Probability of including an extant species in the tree. Known (sampling fraction).
Fossil Sampling Rate ψ (psi) Rate of fossil occurrence per lineage per time unit. Estimated or known from fossil record.
Treatment Probability r (or ω) Probability a sampled extant species is included in the final, trimmed tree (e.g., for pathogen trees). Known (conditioning factor).
Tree Prior — Probability density of the tree under the model. Used for Bayesian inference (e.g., Birth-Death Serial Sampling).
Occurrence Data — Fossil first and last appearance dates. Vector of times for calibration/sampling.

Experimental & Computational Protocols

Protocol 3.1: Implementing the Fossilized Birth-Death (FBD) Model in Bayesian Software

Objective: To estimate speciation, extinction, and fossil sampling rates from a combined dataset of molecular sequences from extant taxa and fossil occurrence dates.

Materials: Time-calibrated phylogenetic tree (or sequence alignment), fossil occurrence table, Bayesian inference software (e.g., BEAST2, RevBayes).

Procedure:

  • Data Preparation:
    • Prepare a NEXUS file containing a starting tree or a molecular sequence alignment for extant taxa.
    • Prepare a fossil occurrence table in a compatible format (e.g., .txt or integrated within the NEXUS). Each fossil should have a taxon name and a minimum (max) and maximum (min) age constraint for its stratigraphic range.
  • Model Specification (BEAST2 Example):
    • Select the Fossilized Birth-Death tree prior.
    • Set parameter priors: Use a log-normal or exponential prior for speciation (λ), extinction (μ), and fossil sampling (ψ) rates. Use a Beta prior for the extant sampling fraction (ρ) if it is not known precisely.
    • Link the fossil occurrences to the tree by specifying which tip or internal node each fossil calibrates.
  • MCMC Setup:
    • Configure the Markov Chain Monte Carlo (MCMC) chain length (e.g., 10-100 million steps), logging frequency, and number of independent runs.
    • Enable sampling of the tree ancestry (the "tree including fossils") to visualize the complete sampled history.
  • Execution & Diagnostics:
    • Run the MCMC analysis.
    • Assess convergence using ESS (Effective Sample Size) values >200 for all parameters in tracer analysis software.
    • Combine and summarize posterior tree distributions using TreeAnnotator, generating a Maximum Clade Credibility tree that includes fossil lineages.

Protocol 3.2: Correcting for Incomplete Taxon Sampling in Macroevolutionary Rate Analyses

Objective: To compute diversification rates from an incomplete phylogeny using sampling fractions.

Materials: Ultrametric phylogeny of sampled species, total species count for clades of interest.

Procedure:

  • Calculate Sampling Fractions:
    • For each clade c in the analysis (e.g., family, order), determine Nc, the number of species included in the phylogeny, and Tc, the total known extant species count for that clade.
    • Compute the clade-specific sampling fraction: ρc = Nc / Tc*.
  • Integrate into Birth-Death Model:
    • In models like the State-Dependent Speciation and Extinction (SSE) models (e.g., implemented in secsse or diversitree in R), specify the sampling.f argument as the vector of ρc* values corresponding to the states/traits at the tree tips.
    • This conditions the likelihood calculation on the fact that only a fraction of species from each state/clade were sampled, preventing biased rate estimates where higher-diversity clades are undersampled.
  • Model Fitting & Comparison:
    • Fit the SSE model with the corrected sampling fractions.
    • Compare its fit (via AIC or likelihood ratio test) to a model assuming uniform or complete sampling to quantify the bias correction's impact.

Visualizations

sampling_bias_workflow TrueHistory True Evolutionary History (Speciation & Extinction Events) SamplingProcess Sampling Process (ρ for extant, ψ for fossils) TrueHistory->SamplingProcess Governs ObservedData Observed Data (Incomplete Phylogeny + Fossils) SamplingProcess->ObservedData Produces BDModel Birth-Death Model (λ, μ, ρ, ψ parameters) ObservedData->BDModel Input for Inference Bayesian MCMC Inference BDModel->Inference Defines Likelihood Posterior Posterior Estimates (λ, μ, ψ, Complete History) Inference->Posterior Yields Posterior->TrueHistory Infers

Title: Workflow for Correcting Sampling Bias in Phylogenetics

fbd_tree_concept cluster_true True Lineages (Inferred) T1 T2 T1->T2 T3 T1->T3 T4 T2->T4 T5 Sampled Extant Tip T2->T5 T6 Sampled Fossil T3->T6 T7 T3->T7 Process Sampling Process: ρ (Extant) & ψ (Fossil) Obs Sampled Extant Sampled Fossil Process->Obs Filters

Title: Fossilized Birth-Death (FBD) Tree Sampling Concept

The Scientist's Toolkit

Table 2: Key Research Reagent Solutions for Sampling-Bias Aware Analysis

Item Function/Description Example Software/Package
Bayesian Evolutionary Analysis Tool Primary platform for Bayesian phylogenetic analysis with integrated tree priors for incomplete sampling. BEAST2 (with SA and FBD packages)
Probabilistic Programming Framework Flexible platform for specifying custom birth-death models with complex sampling scenarios. RevBayes
R Phylogenetics Suite Environment for implementing SSE models, fitting birth-death models with sampling fractions, and visualizing results. R packages: TreeSim, ape, phytools, diversitree, secsse
Tree Simulation Software Generates synthetic phylogenies under defined birth-death-sampling parameters for testing and validation. TreeSim (R), pyvolve (Python)
Fossil Data Handling Library Manages, cleans, and formats fossil occurrence data for integration with phylogenetic analyses. palaeoverse (R), paleotree (R)
MCMC Diagnostics Tool Assesses convergence and mixing of Bayesian MCMC runs, essential for validating parameter estimates. Tracer
Tree Visualization & Annotation Annotates posterior tree sets with node ages and posterior supports, and creates publication-quality figures. FigTree, ggtree (R)

Application Notes & Protocols (Framed within a Thesis on Bayesian Birth-Death Analysis for Diversity History Research)

In Bayesian birth-death (BD) modeling of species diversification, model selection is critical. Overparameterized models (e.g., with many episodic rate shifts) can overfit noisy fossil or molecular data, obscuring true macroevolutionary signals. Bayesian Model Averaging (BMA) and explicit hypothesis testing via Bayes Factors offer a principled framework to balance complexity. BMA integrates over model uncertainty by weighting the posterior distribution of parameters (e.g., speciation/extinction rates) by the posterior model probabilities. This avoids the "winner's curse" of selecting a single, potentially overfitted model. This document provides protocols for implementing these techniques within a broader thesis analyzing historical biodiversity dynamics, with applications in identifying genuine diversification rate shifts relevant to understanding historical drug discovery sources.

Quantitative Comparison of Model Selection Methods

Table 1: Core Methods for Balancing Model Complexity in Bayesian Birth-Death Analysis

Method Key Metric / Output Quantitative Criterion Advantage for BD Analysis Disadvantage
Bayesian Model Averaging (BMA) Posterior Model Probability (PMP) PMP = (Marginal Likelihood * Prior Model Prob) / Total Evidence Propagates model uncertainty into parameter estimates (e.g., net diversification). Robust. Computationally intensive. Requires defining a meaningful model space.
Bayes Factor (BF) Testing Bayes Factor (BF₁₂) BF₁₂ = Marginal Likelihood(M₁) / Marginal Likelihood(M₂). BF > 10 strong for M₁. Directly tests nested hypotheses (e.g., constant vs. 1-shift BD model). Sensitive to prior choices on parameters. Calculation of marginal likelihoods can be tricky.
Reversible Jump MCMC (RJMCMC) Model Indicator Variable Samples model space alongside parameters. Inferred model = most frequently visited. Automatically explores model complexity. Ideal for unknown number of rate shifts. Complex to implement and diagnose convergence.
Widely Applicable Information Criterion (WAIC) WAIC Score WAIC = -2 * (log pointwise predictive density - penalty for effective parameters). Lower = better. Approximates cross-validation. Computed from posterior samples. Not fully Bayesian; can be unstable with weak data.

Table 2: Example Output from a Birth-Death Model Comparison Study (Synthetic Data)

Model (Num. of Rate Shifts) Log Marginal Likelihood (lnℤ) Bayes Factor (vs. Constant) Posterior Model Probability Estimated Parameters
Constant Rates (M0) -245.3 1 (Reference) 0.02 λ=0.2, μ=0.1
1 Shift (M1) -238.1 735.7 (Strong for M1) 0.78 λ₁=0.15, λ₂=0.25, μ=0.09
2 Shifts (M2) -239.8 73.7 (Positive for M2) 0.20 λ₁=0.14, λ₂=0.3, λ₃=0.18, μ=0.1

Experimental Protocols

Protocol 3.1: Bayesian Model Averaging for Diversification Rates

Objective: To estimate the probability-averaged speciation rate through time from a set of candidate birth-death models. Inputs: Time-calibrated phylogenetic tree (fossil-extended or molecular); set of candidate BD models (e.g., constant rates, 1-3 episodic shifts). Software: BEAST2, RevBayes, or R (package birthdeath or RPANDA).

Procedure:

  • Model Specification: Define K candidate BD models (M1...Mk) with varying complexity (e.g., different numbers of rate shift times).
  • Prior Definition: Assign equal prior probability to each model (1/K). Set plausible priors for parameters (e.g., speciation λ ~ Γ(2,10)).
  • Marginal Likelihood Estimation: For each model, run an MCMC chain (min 10^7 steps). Compute log marginal likelihood (lnℤ) using:
    • Path Sampling/Stepping Stone Sampling: Recommended. Use 50-100 power posteriors, α ~ Beta(0.3).
    • Harmonic Mean Estimator: Avoid; known instability.
  • Calculate Posterior Model Probabilities (PMP): PMP(M_i) = exp(lnℤ_i) * Prior(M_i) / Σ[exp(lnℤ_k) * Prior(M_k)]
  • Parameter Averaging: For parameter of interest (e.g., net diversification at time t), compute: E[rate|data] = Σ (PMP_i * E[rate|data, M_i]) where E[rate|data, M_i] is the posterior mean under model i.
  • Validation: Check effective sample size (ESS > 200) for all parameters and likelihoods across runs.

Protocol 3.2: Bayes Factor Testing for a Rate Shift Hypothesis

Objective: Test if a mass extinction event at time t significantly improved model fit. Inputs: Phylogenetic tree; fossil occurrence data around time t.

Procedure:

  • Define Nested Models: M0: Constant-rate BD model. M1: BD model with a mass extinction (instantaneous loss of a fraction of species) at time t.
  • MCMC & Marginal Likelihood: Run Path Sampling for both models as in Protocol 3.1, steps 3-4.
  • Compute Bayes Factor: 2lnBF₁₀ = 2 * (lnℤ(M1) - lnℤ(M0)) Interpret: 2lnBF > 6 = "Strong" evidence for mass extinction model M1.
  • Sensitivity Analysis: Vary the prior on the extinction fraction (e.g., Uniform(0,0.8) vs. Beta(2,2)) to assess robustness of BF.

Visualization of Methodologies

Diagram 1: Bayesian Model Averaging Workflow for BD Models

bma_workflow Data Time Tree & Fossil Data ModelSpace Define Model Space (M0, M1...Mk) Data->ModelSpace PriorSpec Assign Priors (Models & Parameters) ModelSpace->PriorSpec MCMCRuns Run MCMC for Each Model PriorSpec->MCMCRuns CalcML Calculate Marginal Likelihoods (lnZ) MCMCRuns->CalcML CalcPMP Compute Posterior Model Probabilities (PMP) CalcML->CalcPMP AvgParams Average Parameters Weighted by PMP CalcPMP->AvgParams Output Model-Averaged Diversification Curve AvgParams->Output

Title: BMA Workflow for Birth-Death Models

Diagram 2: Model Space for Birth-Death Analysis

model_space BDModel Birth-Death Model Class Constant Constant Rates (λ(t), μ(t)) BDModel->Constant Episodic Episodic Shifts (e.g., Mass Extinction) BDModel->Episodic TimeVar Time-Varying (e.g., Logistic) BDModel->TimeVar M0 M0 Episodic->M0 0 shifts M1 M1 Episodic->M1 1 shift M2 M2 Episodic->M2 2 shifts DotDot DotDot Episodic->DotDot ...

Title: Model Space Complexity in BD Analysis

The Scientist's Toolkit: Key Research Reagents & Materials

Table 3: Essential Computational & Data Resources for Bayesian BD Analysis

Item / Solution Function & Relevance to Thesis Example / Specification
Time-Calibrated Phylogeny Primary input data. Represents evolutionary relationships with node ages. Can be molecular (dated with fossils) or fossil-only. BEAST2 .tre output; Fossilized Birth-Death tree.
Fossil Occurrence Database Provides calibration points and direct evidence of past diversity. Critical for constraining extinction times. Paleobiology Database (PBDB) extract; custom fossil dataset.
Bayesian Evolutionary Analysis Software Platform for implementing BD models, MCMC sampling, and marginal likelihood calculation. BEAST2 (with BirthDeathSkyline package), RevBayes.
Marginal Likelihood Estimation Tool Specialized algorithm to compute the model evidence (lnℤ), enabling BMA and BF. Path Sampling/Stepping Stone Sampling in BEAST2 (modelselection package).
High-Performance Computing (HPC) Cluster Essential for computationally intensive MCMC runs across multiple complex models (10^7-10^9 steps). Slurm job array for running multiple models in parallel.
Phylogenetic Data Format Standardized format for exchanging trees and associated data. Nexus file format (.nex) with TREES and DATA blocks.
Priors Database/Repository Curated list of biologically plausible prior distributions for speciation/extinction rates and shift times. Published empirical rates from meta-analyses; fbd.priors file.

Within a thesis on Bayesian birth-death analysis for reconstructing species diversity histories from molecular phylogenies, computational speed is a critical bottleneck. Analyses using software like BEAST2 with birth-death skyline models can require weeks of computation on standard workstations. This note details integrated strategies employing the BEAGLE library, cloud computing resources, and novel approximate methods to reduce time-to-result from months to days, enabling more complex model exploration and robust hypothesis testing in evolutionary and pharma-relevant pathogen studies.

Application Notes & Protocols

Leveraging the BEAGLE Library for Phylogenetic Likelihood Acceleration

Protocol: Enabling and Configuring BEAGLE in BEAST2 BEAGLE (Broad-platform Evolutionary Analysis General Likelihood Evaluator) is an open-source library that uses parallel computing on GPUs and multi-core CPUs to accelerate phylogenetic likelihood calculations.

  • Prerequisite Installation:

    • Install the latest version of BEAST2 (v2.7.5+).
    • Install BEAGLE from the official GitHub repository (https://github.com/beagle-dev/beagle-lib). Precompiled binaries are available for macOS and Windows; Linux requires compilation.
    • Verify installation by running beagle-info in the terminal, which should list available computational resources (CPU threads, GPUs).

  • BEAST2 XML Configuration:

    • In your BEAST2 XML input file, ensure the following flags are set within the <run> block:

    • Critical Parameters:

      • useGPU="true": Offloads calculations to the graphics card. Set to false to use CPU-only threading.
      • useDoublePrecision="true": Uses double-precision arithmetic, recommended for stability in Bayesian analysis. For faster, less precise scans, set to false.
      • scaling="always": Prevents underflow in long sequence alignments.

  • Benchmarking Performance:

    • Run an identical analysis (e.g., a birth-death skyline model on a 100-taxon, 10,000bp alignment) with useGPU="false" (CPU) and useGPU="true" (GPU).
    • Measure the time in seconds per million MCMC steps. Record results in Table 1.

Table 1: BEAGLE Acceleration Benchmark

Hardware Configuration BEAGLE Flags Time per 1M Steps (sec) Relative Speedup
Intel Xeon 8-core CPU useGPU="false", threads=8 2,850 1.0x (Baseline)
NVIDIA Tesla V100 GPU useGPU="true", precision=double 312 9.1x
NVIDIA RTX A5000 GPU useGPU="true", precision=single 198 14.4x

Cloud Computing Orchestration for High-Throughput Analysis

Protocol: Deploying BEAST2 Analyses on AWS Batch Cloud platforms allow scalable, parallel runs of multiple independent MCMC chains or model comparisons.

  • Containerization:

    • Create a Dockerfile that installs BEAST2, BEAGLE, and all necessary packages.

    • Build and push the image to Amazon ECR.

  • Workflow Orchestration with Nextflow:

    • Write a main.nf Nextflow script to manage job submission to AWS Batch.

    • This script automatically dispatches jobs to GPU or CPU instances based on model complexity.

  • Cost-Performance Optimization:

    • Select cloud instances based on the BEAGLE benchmark. Use GPU instances (e.g., AWS g4dn or p3 series) for demanding relaxed-clock models, and cheaper CPU instances (e.g., c5 series) for simpler strict-clock models.
    • Use spot instances for non-urgent, fault-tolerant runs to reduce costs by 60-90%.

Table 2: Cloud Instance Cost-Performance Analysis

AWS Instance Type vCPUs GPU Approx. Cost per Hour Recommended Use Case
c5.4xlarge 16 None $0.68 Strict-clock birth-death, posterior sampling.
g4dn.xlarge 4 1 T4 Tensor Core $0.526 Relaxed-clock models, moderate dataset size.
p3.2xlarge 8 1 V100 Tensor Core $3.06 Large phylogenies (>500 taxa), complex skyline models.

Incorporating Approximate Methods for Model Exploration

Protocol: Using Path Sampling/Stepping Stone Sampling for Marginal Likelihood Estimation Approximate methods like Path Sampling (PS) and Stepping Stone Sampling (SS) allow rapid comparison of competing birth-death models (e.g., constant rate vs. skyline) to select the best-fitting model before full, lengthy MCMC runs.

  • Experimental Design:

    • Prepare BEAST2 XML files for at least two competing models (e.g., Birth-Death (BD) vs. Birth-Death Skyline (BDSKY)).
    • For each model, run a preliminary, short MCMC chain (length=1,000,000) to generate a sample from the posterior.

  • Marginal Likelihood Calculation Protocol:

    • Using the BEAST2.app package modelselection (v1.0+):

    • Repeat for the competing model.

  • Model Selection Decision:

    • Calculate the log Bayes Factor (BF): 2 * (log Marginal Likelihood Model A - log Marginal Likelihood Model B).
    • Interpret: BF > 10 provides "very strong" support for Model A over Model B. Only proceed with full, long MCMC analysis on the strongly supported model.

The Scientist's Toolkit: Research Reagent Solutions

Item Function in Bayesian Birth-Death Analysis
BEAST2 Software Package Core platform for Bayesian phylogenetic analysis, implementing birth-death skyline and related models.
BEAGLE Library High-performance computational library that accelerates likelihood calculations on GPU/CPU hardware.
CIPRES Science Gateway / XSEDE Web-based portal for submitting jobs to high-performance computing clusters without local setup.
Amazon EC2 / Google Cloud VMs Scalable cloud virtual machines for on-demand, parallel computation of multiple MCMC chains.
Docker Containers Reproducible, portable environments encapsulating BEAST2, BEAGLE, and all dependencies.
Nextflow / Snakemake Workflow managers to orchestrate complex, multi-step phylogenetic analyses on cloud/cluster infrastructure.
Tracer Software For diagnosing MCMC convergence, analyzing effective sample sizes (ESS), and summarizing parameter estimates.
TreeAnnotator Generates a maximum clade credibility tree from the posterior tree distribution.
FigTree / IcyTree Visualization tools for displaying time-scaled phylogenetic trees with node bars representing uncertainty.

Visualizations

G Start Input Data: Time-scaled Phylogeny & Taxon Data M1 Model Exploration (Approximate Methods) Start->M1 PS/SS Sampling M2 Full MCMC Analysis (BEAGLE Accelerated) M1->M2 Select Best Model (Bayes Factor > 10) M3 Cloud-based Parallel Execution M2->M3 Dispatch Multiple Chains End Posterior Distribution of Diversification Rates M3->End Combine & Summarize

Diagram Title: Computational Acceleration Workflow for Bayesian Birth-Death Analysis

G Sub Phylogenetic Substitution Model LikelihoodCalc Likelihood Calculation (Computational Bottleneck) Sub->LikelihoodCalc Clock Molecular Clock Model (Strict/Relaxed) Clock->LikelihoodCalc TreePrior Birth-Death Tree Prior (e.g., Skyline Model) TreePrior->LikelihoodCalc BEAGLE BEAGLE Library (GPU/CPU Offload) LikelihoodCalc->BEAGLE Accelerates Posterior Joint Posterior Distribution P(Parameters, Tree | Data) LikelihoodCalc->Posterior

Diagram Title: BEAGLE's Role in the Bayesian Phylogenetic Pipeline

Bayesian Birth-Death vs. Alternative Methods: Validating Inferences and Choosing the Right Tool

Within the context of a broader thesis on Bayesian birth-death analysis for diversity history research, understanding the alternative statistical paradigms is crucial. This document provides application notes and experimental protocols for comparing Frequentist and Machine Learning (ML)-based Maximum Likelihood (ML) approaches, such as the TreePar package, for inferring diversification rates from phylogenetic trees.

Quantitative Comparison of Methodological Paradigms

Table 1: Core Comparison of Bayesian, Frequentist ML, and ML-based (e.g., TreePar) Approaches

Feature Bayesian Birth-Death (Reference) Frequentist Maximum Likelihood (e.g., ape, diversitree) ML-based with Shifts (TreePar)
Primary Output Posterior distribution of parameters (rates, shift times). Point estimates (MLEs) with confidence intervals. Point estimates of MLEs for rates and shift times.
Uncertainty Quantification Intrinsic (credible intervals from posterior). Asymptotic (confidence intervals via Hessian). Bootstrap confidence intervals for rates and shift times.
Handling Complexity Excellent; prior distributions regularize complex models. Can overfit with many parameters; model selection (AIC, BIC) required. Explicitly models discrete rate shifts over time; penalized likelihood controls complexity.
Computational Demand High (MCMC sampling). Low to Moderate. Moderate (numerical optimization over shift times).
Prior Information Incorporated directly via priors. Not incorporated. Not incorporated in standard form.
Temporal Rate Variation Continuous (e.g., skyline) or discrete shifts. Typically constant or simple functions. Core Strength: Models discrete rate shifts at specific times.
Interpretability Full probabilistic interpretation. Likelihood-based, intuitive. Direct inference of when diversification regime changed.

Table 2: Example Performance Metrics on Simulated Phylogenetic Data (Hypothetical data based on published benchmarks)

Simulation Scenario Method Mean Error in Shift Time (Myr) Power to Detect Shift (%) False Positive Rate (%)
Single Strong Shift TreePar 1.2 98 5
Bayesian Skyline 3.5 95 8
Constant Rate ML N/A 0 10
Multiple Weak Shifts TreePar 5.8 75 12
Bayesian Skyline 4.1 80 15
Constant Rate ML N/A 0 8

Experimental Protocol: Analyzing Diversification Shifts with TreePar

Protocol Title: Inferring Time-Dependent Diversification Rate Shifts from a Dated Phylogeny Using TreePar.

Objective: To identify significant changes (shifts) in speciation and/or extinction rates through time using maximum likelihood optimization.

Materials & Input Data:

  • A time-calibrated phylogenetic tree (ultrametric) in Newick format.
  • The R statistical environment (v4.0+).
  • Installed R packages: TreePar, ape, phytools.

Procedure:

Step 1: Data Preparation and Import

Step 2: Setting Analysis Parameters

Step 3: Likelihood Optimization and Model Selection

Step 4: Statistical Testing and Shift Time Inference

Step 5: Bootstrap Analysis for Confidence Intervals (Critical)

Visualization of Methodological Workflows

G Start Input: Ultrametric Phylogenetic Tree Preprocess Tree Preprocessing (Resolve polytomies, check binary) Start->Preprocess ParamGrid Define Time Grid & Max Shift Number Preprocess->ParamGrid Optimize Penalized ML Optimization (TreePar core) ParamGrid->Optimize ModelSelect Model Selection (AIC / Likelihood Ratio Test) Optimize->ModelSelect OutputMLE Output: MLEs of Shift Times & Rates ModelSelect->OutputMLE Bootstrap Bootstrap Resampling for Confidence Intervals OutputMLE->Bootstrap FinalResult Final Inference: Significant Shifts with Uncertainty Bootstrap->FinalResult

Workflow for TreePar-based Diversification Shift Analysis

H Past Past (t=10) Shift Rate Shift (t=5) Past->Shift λ1, μ1 Present Present (t=0) Shift->Present λ2, μ2

TreePar Models Discrete Rate Shifts Over Time

Table 3: Essential Computational Tools for Diversification Rate Analysis

Item / Resource Function / Purpose Example / Source
Ultrametric Phylogeny Time-calibrated input tree for all rate analyses. Output from BEAST2, treePL, or chronos.
R Environment Statistical computing platform for analysis. https://www.r-project.org/
TreePar R Package Implements ML-based birth-death models with rate shifts. CRAN repository: install.packages("TreePar")
ape R Package Core package for reading, writing, and plotting phylogenies. CRAN: install.packages("ape")
diversitree R Package Frequentist ML analysis for comparative diversification. CRAN: install.packages("diversitree")
BEAST2 / RevBayes Bayesian phylogenetic inference & birth-death analysis. https://www.beast2.org/
TreeSim R Package Simulate phylogenetic trees under birth-death models for testing. CRAN: install.packages("TreeSim")
High-Performance Computing (HPC) Cluster For computationally intensive bootstrap or Bayesian MCMC runs. Institutional or cloud-based (AWS, Azure).

This document provides application notes and protocols for selecting and implementing population genetic models within a Bayesian framework for diversity history research. The central thesis posits that the birth-death skyline (BDS) model offers a more flexible and biologically interpretable prior for times-calibrated phylogenetic inference in many epidemic and macroevolutionary contexts compared to the coalescent.

Foundational Model Comparison

Table 1: Core Model Comparison – Birth-Death Skyline vs. Coalescent

Feature Birth-Death Skyline (BDS) Prior Coalescent Prior (e.g., Skyline)
Fundamental Process Forward-in-time: models speciation/birth (λ) and extinction/death (μ) rates. Backward-in-time: models the probability of ancestral lineages merging (coalescing).
Primary Data Times of speciation/transmission events (internal nodes) and sampling events (tips). Time to the Most Recent Common Ancestor (TMRCA) and intervals between coalescent events.
Key Parameters Net diversification (λ - μ), turnover (μ/λ), sampling proportion. Effective population size (Ne), growth rate.
Sampling Model Explicitly models serially sampled tips through time (important for viruses, fossils). Typically assumes contemporaneous sampling or simple sampling models.
Inferred Quantities Times of origin, reproductive number (R), rate of becoming non-infectious. Changes in effective population size (Ne) over time.
Best For Epidemics (e.g., HIV, SARS-CoV-2), fossil data, clade diversification with serial sampling. Intra-species population genetics, shallow datasets with contemporaneous sampling.

Protocol: Decision Framework and Benchmarking Workflow

Step-by-Step Model Selection Protocol

Objective: To empirically determine whether a Birth-Death or Coalescent prior is more appropriate for a given time-stamped phylogenetic dataset (e.g., viral sequences, species phylogenies with fossils).

Materials & Input Data:

  • Time-scaled molecular sequence alignment (e.g., .fasta, .nexus).
  • Sample collection dates for each tip.
  • Computational resources (high-performance computing cluster recommended).
  • Software: BEAST 2.xx, TreeAnnotator, Tracer, R with ggtree, coda.

Procedure:

  • Data Preparation & Model Specification:

    • Create two identical XML files in BEAUti (BEAST 2), differing only in the tree prior.
    • File A: Specify a Birth-Death Skyline Contemporary prior. Set appropriate priors for the reproductive number (R) and the becoming-non-infectious rate (δ). Enable the skyline model to allow rates to change over time intervals.
    • File B: Specify a Coalescent Bayesian Skyline prior. Set number of population size groups (e.g., 5-10).
    • For both, use the same clock model (e.g., Relaxed Clock Log Normal) and site model (e.g., HKY+Γ).

  • Bayesian MCMC Execution:

    • Run each analysis (A & B) for an adequate number of steps (e.g., 50-100 million) to achieve ESS > 200 for all key parameters.
    • Perform at least two independent runs per model to assess convergence.

  • Model Comparison via Path Sampling/Stepping Stone Sampling:

    • Using the same model setups, perform Path Sampling (PS) or Stepping Stone Sampling (SS) to estimate the marginal likelihood for Model A (BDS) and Model B (Coalescent).
    • Protocol for Path Sampling in BEAST 2:
      • Use the -p flag with a specified number of steps (e.g., 100-200).
      • Create a series of power posteriors from the prior (beta=0) to the posterior (beta=1).
      • The log marginal likelihood is estimated by numerically integrating over these steps. A difference in log marginal likelihood > 3 is considered positive support for the better-fitting model.

  • Comparative Output Analysis:

    • Compare the estimated effective population size (Ne) trajectory from the coalescent with the estimated reproductive number (R) trajectory from the BDS.
    • Assess the congruence of the maximum clade credibility (MCC) tree topology and node ages between runs.
    • Use posterior predictive simulations to check which model better recapitulates the distribution of node ages and tree shapes in the empirical data.

Table 2: Diagnostic Indicators for Model Preference

Observation Favors Birth-Death Model Favors Coalescent Model
Sampling Scheme Intensive, serial sampling through time. Single, contemporaneous sampling event.
Primary Question When did the epidemic originate? What is R(t)? How has Ne fluctuated over time?
Marginal Likelihood Significantly higher log ML for BDS. Significantly higher log ML for Coalescent.
Tree Prior Saturation Many lineages sampled near the present. Few lineages, deep coalescence.

Visualization: Model Selection and Benchmarking Workflow

G Start Input: Time-Stamped Sequence Data Q1 Primary Aim: Epidemic Dynamics (R0, Origin Time)? Start->Q1 Q2 Sampling Serial & Structured? Q1->Q2 Yes Q3 Primary Aim: Pop. Size History (Ne)? Q1->Q3 No ModBD Use Birth-Death Skyline Prior Q2->ModBD Yes Bench Benchmark: Run Both Models Compare via PS/SS Q2->Bench Uncertain ModCoal Use Coalescent Skyline Prior Q3->ModCoal Yes Q3->Bench Uncertain Analyze Analyze Posteriors & Trajectories ModBD->Analyze ModCoal->Analyze Bench->Analyze

Title: Decision Workflow: Birth-Death vs. Coalescent Prior Selection

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Software & Analytical Tools

Item Category Function & Application Note
BEAST 2 / BEAUti Software Package Core platform for Bayesian evolutionary analysis. Use to set up Birth-Death and Coalescent models, clock models, and run MCMC.
Tracer Diagnostics Tool Visualize MCMC traces, assess convergence (ESS > 200), and compare parameter posterior distributions between models.
TreeAnnotator Tree Tool Generate Maximum Clade Credibility (MCC) trees from posterior tree sets for each model.
PathSampler/StarBEAST2 BEAST 2 Package Perform Path Sampling to calculate marginal likelihoods for rigorous statistical model comparison.
ggtree (R package) Visualization Plot MCC trees with node bars, map trait evolution, and visualize skyline plots (R(t) or Ne(t)) from BEAST outputs.
Skyride Plot Generator Visualization Script Custom script (often in R) to generate smoothed trajectories of Ne or R from piecewise-constant BEAST skyline outputs.
High-Performance Compute Cluster Infrastructure Essential for running long MCMC chains (50M+ steps) and computationally intensive Path Sampling analyses.

Within the broader thesis on Bayesian birth-death analysis for reconstructing species diversity history, model selection is paramount. The choice between different birth-death process parameterizations (e.g., constant-rate, time-dependent, diversity-dependent) fundamentally shapes inferred evolutionary trajectories. This protocol details the application of marginal likelihoods—estimated via Path Sampling (PS) and Stepping-Stone Sampling (SS)—and Bayes Factors to achieve robust, quantitative model comparison, moving beyond heuristic goodness-of-fit measures.

Core Theoretical Framework

Marginal Likelihood & Bayes Factor

The marginal likelihood (M) of model i is the probability of the observed data D given the model, integrated over its parameter space Θi with prior *p*(θ|*Mi): *M_i = p(D | M_i) = ∫_{Θ_i} p(D | θ, M_i) p(θ | M_i) dθ

The Bayes Factor (BF) comparing model i to model j is the ratio of their marginal likelihoods: BF_ij = M_i / M_j A BF_ij > 1 favors model i; thresholds for evidence strength are given in Table 1.

Estimation Methods

Path Sampling (PS): Also known as thermodynamic integration, PS computes the log marginal likelihood by integrating over a path from a reference distribution (e.g., prior) to the posterior. Stepping-Stone Sampling (SS): Uses a series of distributions bridging the prior and posterior, estimating the likelihood ratio via importance sampling.

Application Notes for Birth-Death Analysis

Model Space Definition

In diversity history research, candidate models typically vary in how speciation (λ) and extinction (μ) rates are defined.

  • M1: Constant-Rate Birth-Death (λ, μ constant).
  • M2: Time-Variable Birth-Death (λ(t), μ(t) modeled, e.g., with epoch or skyline models).
  • M3: Diversity-Dependent Birth-Death (λ(N), μ(N)).
  • M4: Environmental-Correlation Models (λ, μ linked to paleoclimate covariates).

Table 1: Bayes Factor Evidence Interpretation

2ln(BF_ij) BF_ij Evidence for Model i
0 to 2 1 to ~3 Not worth more than a bare mention
2 to 6 ~3 to 20 Positive
6 to 10 ~20 to 150 Strong
>10 >150 Very Strong

Note: 2ln(BF) is used for comparison to χ² distribution quantiles.

Table 2: Hypothetical Model Comparison for Cetacean Diversity Dataset

Model Description ln(ML) PS (SE) ln(ML) SS (SE) 2ln(BF) vs. M1 Rank
M2 3-Epoch Variable Rate -125.4 (0.8) -125.1 (0.7) 18.2 1
M4 Temp-Correlated Spec. -129.1 (0.9) -128.8 (0.8) 9.8 2
M1 Constant Rate -134.5 (0.5) -134.2 (0.5) 0.0 3
M3 Linear Diversity-Dep. -135.8 (1.1) -135.6 (1.0) -2.6 4

SE: Standard Error of the estimate. Data is illustrative.

Experimental Protocols

Protocol: Bayesian Birth-Death Model Fitting (Prerequisite)

Purpose: Generate posterior distributions for each candidate model required for PS/SS. Software: BEAST2, RevBayes, or MrBayes with appropriate birth-death packages. Steps:

  • Data Preparation: Format a time-calibrated phylogeny (ultrametric) in Nexus format.
  • Model Specification: Define the prior distributions for all parameters (λ, μ, etc.) for each candidate model M_i.
  • MCMC Setup: Run 2-4 independent Markov Chain Monte Carlo (MCMC) runs per model. Ensure convergence (ESS > 200 for all parameters) using Tracer.
  • Posterior Sampling: Log samples from the stationary posterior distribution. A minimum of 10,000 samples is recommended. Output: Posterior distribution files (.log, .trees) for each model M_i.

Protocol: Marginal Likelihood Estimation via Path Sampling

Purpose: Calculate the log marginal likelihood for a single model. Software: BEAST2 (BEAST package) or custom scripts in R/Python. Steps:

  • Define Path: Create a sequence of K power posteriors (typically K=50-100), where distribution k is defined as p(θ|D, β_k) ∝ p(D|θ)^{β_k} p(θ). The heating schedule {β0=0, ..., βk, ..., β_K=1} is defined, often using a beta(α,1.0) function.
  • Run Power Posterior MCMC: For each k from 0 to K, run a tempered MCMC chain targeting p(θ|D, β_k). Initial state can be from the previous run.
  • Calculate Likelihood: For each run, record the mean of the log likelihood of the samples.
  • Numerical Integration: Compute the log marginal likelihood as: ln(ML_PS) ≈ Σ_{k=1}^{K} ½(β_k - β_{k-1})(\bar{L}_k + \bar{L}_{k-1}) where \bar{L}_k is the mean log-likelihood from run k. Output: ln(ML_PS) with associated standard error.

Protocol: Marginal Likelihood Estimation via Stepping-Stone Sampling

Purpose: Calculate the log marginal likelihood for a single model, often with lower variance than PS. Steps:

  • Define Stepping Stones: Similar to PS, define a sequence of K power posteriors (β0=0,...,βK=1).
  • Sample from Intermediate Distributions: For each step k, draw samples from p(θ|D, β_k). Use importance sampling with the posterior from step k-1 as proposal.
  • Calculate Per-Step Ratio: Compute the likelihood ratio for step k.
  • Aggregate: The marginal likelihood is the product of the K ratios: ML_SS = Π_{k=1}^{K} (1/N Σ_{n=1}^{N} [p(D|θ_{k-1,n})^{β_k - β_{k-1}}]) where θ_{k-1,n} are samples from the (k-1)-th distribution. Output: ln(ML_SS) with associated standard error.

Protocol: Model Comparison via Bayes Factors

Purpose: Rank all candidate models. Steps:

  • Estimate: Obtain ln(ML_i) for each model i using both PS and SS (Protocols 4.2 & 4.3).
  • Compare: For any two models i and j, compute 2ln(BF_ij) = 2[ln(ML_i) - ln(ML_j)].
  • Rank & Interpret: Rank models by highest ln(ML). Use Table 1 to interpret the strength of evidence when comparing the top model to alternatives. Validation: Results from PS and SS should be congruent. Large discrepancies suggest estimation errors.

Visualizations

workflow Start Time-Calibrated Phylogeny M1 Model 1: Constant Rate Start->M1 M2 Model 2: Time-Variable Start->M2 M3 Model 3: Diversity-Dep. Start->M3 M4 Model 4: Env.-Correlated Start->M4 PP1 Power Posterior Sampling (PS/SS) M1->PP1 PP2 Power Posterior Sampling (PS/SS) M2->PP2 PP3 Power Posterior Sampling (PS/SS) M3->PP3 PP4 Power Posterior Sampling (PS/SS) M4->PP4 ML1 ln(ML₁) ± SE PP1->ML1 ML2 ln(ML₂) ± SE PP2->ML2 ML3 ln(ML₃) ± SE PP3->ML3 ML4 ln(ML₄) ± SE PP4->ML4 BF Bayes Factor Calculation & Ranking ML1->BF ML2->BF ML3->BF ML4->BF Result Robust Model Selection BF->Result

Diagram Title: Workflow for Bayesian Birth-Death Model Comparison

BFcalc ML_i ln(MLᵢ) Minus – ML_i->Minus ML_j ln(MLⱼ) ML_j->Minus TimesTwo × 2 Minus->TimesTwo BF_log 2ln(BFᵢⱼ) TimesTwo->BF_log Evidence Evidence Strength BF_log->Evidence

Diagram Title: Bayes Factor Calculation Logic

The Scientist's Toolkit

Table 3: Research Reagent Solutions for Bayesian Birth-Death Model Selection

Item Function / Explanation Example Software/Package
Phylogenetic Inference Platform Core software for specifying models, running MCMC, and sampling posteriors. BEAST2, RevBayes, MrBayes
Marginal Likelihood Estimator Implements Path Sampling and Stepping-Stone Sampling algorithms. BEAST2 (BEAST package), modelselection in RevBayes, nestcheck (Python)
MCMC Diagnostics Tool Assesses convergence, mixing, and effective sample size (ESS) of MCMC runs. Tracer, coda R package
High-Performance Computing (HPC) Resource PS/SS require many (50-100+) independent MCMC runs; parallelization is essential. SLURM cluster, cloud computing (AWS, GCP)
Programming Environment For data wrangling, custom analysis scripts, and visualization. R (tidyverse, ggplot2), Python (pandas, arviz, scipy)
Model Visualization Toolkit Visualizes birth-death rate-through-time and diversity trajectories from posterior. TESS R package, bdvt in RevBayes, pastis
Data Repository Public archive for phylogenetic data and models to ensure reproducibility. Dryad, Figshare, GitHub

Within the broader thesis on Bayesian birth-death analysis for diversity history research, this document presents Application Notes and Protocols for validating estimated speciation and extinction rates against datasets with known underlying histories. Robust validation is critical for applying these models to empirical questions in macroevolution, epidemiology (e.g., viral lineage diversification), and drug development (e.g., cancer cell population dynamics under therapeutic pressure).

Core Data Presentation

The following tables summarize key quantitative findings from validation studies using simulated and empirical benchmark datasets.

Table 1: Performance of Bayesian Birth-Death Models on Simulated Phylogenies

Simulation Scenario (True History) Mean Estimated Speciation Rate (λ) 95% HPD Interval for λ Mean Estimated Extinction Rate (μ) 95% HPD Interval for μ Coverage Probability (True λ in HPD) Model Used (BEAST2/TESS)
Constant-rate Birth-Death (λ=0.1, μ=0.03) 0.102 [0.08, 0.125] 0.031 [0.01, 0.055] 0.94 Birth-Death Skyline
Time-Dependent Rate (Linear Decline in λ) 0.15 (at present) [0.12, 0.18] 0.04 [0.02, 0.07] 0.89 Skyline
Mass Extinction Event (50% loss at t=10) 0.095 [0.07, 0.12] 0.065 [0.04, 0.09] 0.91 Mass-Extinction BD

Table 2: Validation on Empirical Benchmark Datasets (Known from Fossil Records)

Empirical System (Clade) Source (Reference) Inferred Net Diversification Rate (λ - μ) Inferred Turnover (μ/λ) Concordance with Fossil-Based Rate Estimates? (Y/N/Partial) Key Discrepancy Note
Cenozoic Mammals PBDB, PHYLACINE 0.05 per Myr 0.6 Partial Molecular estimates suggest higher extinction rates.
Paleozoic Trilobites Fossil record only 0.02 per Myr 0.8 Y High turnover consistent with fossil data.
Caribbean Reef Corals Fossil & Molecular 0.03 per Myr 0.75 N Molecular phylogenies underestimate extinction.

Experimental Protocols

Protocol 3.1: Simulating Benchmark Phylogenies with Known Parameters

Objective: Generate phylogenetic trees with known speciation (λ) and extinction (μ) histories for method validation.

Materials: High-performance computing cluster, R statistical software (v4.3+).

Reagents/Software: R packages: TreeSim, TESS, ape, phytools.

Procedure:

  • Define Simulation Scenarios: Specify parameters for at least three distinct historical models:
    • Constant-rate birth-death (λ=0.1, μ=0.03, crown age=50 Myr).
    • Time-variable birth-death (λ declines linearly from 0.2 to 0.05 over 50 Myr, μ=0.03).
    • Mass-extinction birth-death (λ=0.1, μ=0.03, 70% species loss at t=30 Myr).
  • Generate Replicates: For each scenario, use TreeSim::sim.bd.age or TESS::tess.sim.taxa to simulate 100 replicate phylogenies.
  • Output Data: Save each resulting Newick tree file and its corresponding true parameter log file.

Protocol 3.2: Bayesian Estimation of Rates from Phylogenetic Data

Objective: Infer speciation and extinction rates from a molecular phylogeny using Bayesian methods.

Materials: Time-calibrated phylogenetic tree (Newick format), BEAST2 software suite (v2.7+).

Reagents/Software: BEAST2, BEAUti, TreeAnnotator, Tracer. Model packages: bdmm, BDSKY, SAVD.

Procedure:

  • Model Specification in BEAUti:
    • Load the tree file.
    • Select the "Birth-Death Skyline Contemporary" model for time-dependent rates.
    • Set priors: Use a log-normal prior for reproductiveNumber (R0 = λ/μ) and a uniform prior for becomeUninfectiousRate (μ).
    • Set up skyline: Define 5-10 time intervals for rate changes.
    • Set MCMC chain length to 10-50 million steps, logging parameters every 10,000 steps.
  • Run MCMC: Execute BEAST2 with the generated XML file.
  • Diagnostics & Summarization:
    • Use Tracer to assess ESS (>200) and convergence.
    • Use TreeAnnotator to generate a maximum clade credibility tree, discarding initial 10% as burn-in.
  • Output: Summarized rates (mean, HPD) through time and an annotated tree.

Protocol 3.3: Cross-Validation with Fossil Time Series Data

Objective: Compare Bayesian birth-death estimates to independent rate estimates from the fossil record.

Materials: Fossil occurrence dataset (from Paleobiology Database), per-taxon molecular phylogeny for the same clade.

Reagents/Software: R packages: paleotree, divDyn, ggplot2.

Procedure:

  • Process Fossil Data: Use paleotree::bin_timeData to bin fossil occurrences into 1-5 Myr intervals. Calculate per-interval origination and extinction rates using divDyn::divDyn.
  • Align Timescales: Temporally align the fossil rate curve with the time-varying rate estimates from Protocol 3.2.
  • Statistical Comparison: Calculate correlation coefficients (Pearson's r) between the fossil-derived diversification rate (origination - extinction) and the molecularly inferred net diversification rate (λ - μ) through time.
  • Identify Discrepancies: Note time intervals where 95% HPDs from molecular analysis do not overlap with fossil-derived point estimates.

Mandatory Visualizations

G Start Start: Research Objective Sim Simulate Phylogenies (Protocol 3.1) Start->Sim Emp Curate Empirical Benchmark Data Start->Emp Bay Bayesian Rate Estimation (Protocol 3.2) Sim->Bay Emp->Bay Foss Fossil Data Analysis (Protocol 3.3) Emp->Foss Val Validation & Comparison Bay->Val Foss->Val End Output: Validated Rate Estimates Val->End

Title: Workflow for Validating Birth-Death Rate Estimates

G Truth Known History (True λ, μ, Events) Data1 Simulated Phylogeny Truth->Data1 Generative Process Comp1 Compare: λ*, μ* vs True λ, μ Truth->Comp1 Model Bayesian Birth-Death Model Data1->Model Data2 Empirical Phylogeny + Fossil Timeseries Data2->Model Inf1 Inferred Rates (λ*, μ*) with HPDs Model->Inf1 Inf2 Inferred Rates (λ†, μ†) with HPDs Model->Inf2 Inf1->Comp1 Comp2 Compare: λ†, μ† vs Fossil Rates Inf2->Comp2 Eval Evaluation: Coverage, Bias, Accuracy Comp1->Eval Comp2->Eval

Title: Logical Framework for Rate Estimate Validation

The Scientist's Toolkit: Research Reagent Solutions

Item/Category Primary Function in Validation Pipeline Example Product/Software
Phylogeny Simulation Package Generates trees under known birth-death processes for testing model accuracy. TreeSim (R), TESS (R), DendroPy (Python)
Bayesian MCMC Platform Engine for statistical inference of parameters from phylogenetic data. BEAST2, RevBayes, MrBayes
Birth-Death Model Library Implements specific birth-death likelihood calculations for MCMC. BEAST2 packages: BDSKY, bdmm, SAVD
Fossil Data Analysis Tool Calculates origination/extinction rates from fossil occurrence tables. paleotree (R), divDyn (R)
MCMC Diagnostics Tool Assesses convergence and mixing of Bayesian runs. Tracer, coda (R), arviz (Python)
High-Performance Computing (HPC) Resource Enables long MCMC chains and large simulation replicates. SLURM cluster, cloud computing (AWS, GCP)

Application Notes

Birth-death stochastic processes are foundational models for inferring rates of lineage origination (birth, λ) and extinction (death, μ) from phylogenetic trees. Their integration with direct observational data from paleontology (fossil occurrences) or epidemiology (case counts) significantly enhances the robustness and interpretability of diversity dynamics. Within a Bayesian framework, this synthesis allows for the joint estimation of parameters while rigorously quantifying uncertainty, reconciling model-based inference with empirical evidence.

Key Applications:

  • Paleobiology: Calibrating molecular clock analyses with the fossil record to estimate divergence times and absolute diversification rates. Testing hypotheses about mass extinction events, radiations, and the impact of environmental covariates.
  • Epidemiology (Phylodynamics): Estimating the basic reproduction number (R₀) and effective population size through time from pathogen genetic sequences, validated against epidemiological incidence data.
  • Drug & Intervention Development: Modeling the emergence of drug-resistant lineages under selective pressure, integrating genetic data with clinical trial outcomes to forecast efficacy timelines.

Quantitative Data Synthesis

Table 1: Core Parameters in Integrated Birth-Death Analyses

Parameter Typical Symbol Paleontological Context Epidemiological Context Data Sources for Integration
Speciation/Birth Rate λ Rate of new species formation Infection rate, transmission rate (β) Phylogenetic branch lengths, fossil first appearances, incidence curve
Extinction/Death Rate μ Species extinction rate Recovery/Death rate (δ), removal rate Fossil last appearances, stratigraphic ranges, case removal data
Net Diversification r = λ - μ Net diversity growth Epidemic growth rate Counts of taxa/cases through time
Reproduction Number R₀ = λ / μ – Average secondary cases Serial interval & growth rate, lineage branching patterns
Sampling Proportion ρ / ψ Fraction of fossils preserved & discovered Fraction of cases sequenced/reported Collection bias, surveillance intensity metrics

Table 2: Comparison of Data Integration Approaches

Approach Method Key Advantage Primary Challenge
Fossilized Birth-Death (FBD) Fossils as direct ancestors or sampled ancestors in the tree. Uses fossil occurrence data directly within tree model. Requires precise fossil dating & taxonomic assignment.
Node Dating & SABD Fossil calibrations as node priors (e.g., uniform, skewed normal). Flexible; uses fossil-based minimum/maximum ages. Often treats fossil data separately from branching process.
Birth-Death Skyline Models Time-varying rates estimated from phylogeny, compared to external time series. Identifies rate shifts through time (e.g., mass extinction). Separates phylogenetic inference from external data comparison.
Structured Epidemiological Models Incidence data informs prior distributions for birth-death parameters. Epidemiological dynamics directly inform tree prior. Complex coupling requires bespoke model development.

Experimental Protocols

Protocol 1: Integrating Fossil Occurrence Data with Phylogenetic Inference using the Fossilized Birth-Death (FBD) Model

Objective: To infer a time-calibrated phylogeny and estimate speciation, extinction, and fossil sampling rates from combined molecular and morphological/fossil data.

Materials: See "The Scientist's Toolkit" below.

Workflow:

  • Data Curation: Compile a molecular sequence alignment for extant taxa and a morphological character matrix for both extant and fossil taxa. Prepare a fossil occurrence file with minimum and maximum ages for each fossil specimen.
  • Tree Model Specification: In Bayesian software (e.g., BEAST2, RevBayes), specify the FBD process as the tree prior. Set parameters for:
    • Diversification rate (λ - μ).
    • Turnover (μ/λ).
    • Fossil sampling proportion (ψ).
  • Tip Dating: Include fossil taxa as tips in the analysis, with ages drawn from uniform distributions bounded by their stratigraphic age ranges.
  • Clock & Substitution Models: Apply appropriate molecular clock (e.g., relaxed lognormal) and nucleotide substitution models. For combined analysis, use a clock model for the morphological partition.
  • MCMC Execution: Run a Markov Chain Monte Carlo (MCMC) analysis for sufficient generations (e.g., 50-100 million), sampling every 10,000 steps.
  • Validation & Analysis: Assess MCMC convergence (effective sample size > 200). Summarize the maximum clade credibility tree. Extract posterior estimates for λ, μ, ψ, and tree topology.

Protocol 2: Phylodynamic Analysis of an Epidemic with Integrated Case Count Data

Objective: To estimate the effective reproduction number (Rₑ(t)) through time from a pathogen phylogeny, using empirical incidence data to inform the clock or population size model.

Materials: See "The Scientist's Toolkit" below.

Workflow:

  • Sequence & Data Alignment: Assemble a dataset of pathogen genome sequences with exact collection dates. Align sequences and curate a matching time series of reported case counts or incidence.
  • Phylogenetic Reconstruction: Use a coalescent or birth-death skyline model as a tree prior in BEAST2. Apply a strict or relaxed molecular clock model.
  • Data Integration via Gaussian Random Walk:
    • Model the log of the effective reproduction number (log(Rₑ(t))) as a Gaussian Random Walk through time.
    • Use the incidence data to construct a directed prior on the number of new infections through time. The expected incidence at time t is proportional to the product of Rₑ(t) and the number of susceptible individuals (inferred or modeled).
  • Model Configuration: Link the birth rate (λ) in the birth-death model to Rₑ(t) and the removal rate (e.g., λ(t) = Rₑ(t) * μ). The removal rate (μ) can be informed by known average infection duration.
  • MCMC Execution & Skyline Plot: Perform Bayesian inference. Generate a skyline plot of the posterior median and credible intervals for Rₑ(t) through time, overlaying the raw incidence data for visual validation.

Mandatory Visualizations

FBD_Workflow A 1. Data Curation B Molecular Sequences (Extant Taxa) A->B C Morphological Matrix (Extant + Fossils) A->C D Fossil Occurrence Data (Ages, Uncertainty) A->D E 2. Model Specification (FBD Tree Prior) B->E C->E D->E F 3. Tip Dating (Fossils as Tips) E->F G 4. Bayesian MCMC (BEAST2/RevBayes) F->G H 5. Posterior Output G->H I Time-Calibrated Phylogeny H->I J Posterior Estimates: λ, μ, ψ, Topology H->J

Fossilized Birth Death Analysis Protocol

Phylo_Epi_Integration Data Input Data Seq Time-Stamped Pathogen Sequences Data->Seq Inc Epidemiological Incidence Curve Data->Inc BD Birth-Death Skyline Prior Seq->BD Clock Molecular Clock Model Seq->Clock InfPrior Informed Prior on New Infections Inc->InfPrior Model Integrated Model Output Output Model->Output BD->Model Link Coupling: λ(t) = Rₑ(t) * μ BD->Link GRW Gaussian Random Walk on log(Rₑ(t)) GRW->Model GRW->Link InfPrior->Model R_t Estimated Rₑ(t) through time Output->R_t Tree Time-Scaled Phylogeny Output->Tree

Integrating Incidence Data into Phylodynamics

The Scientist's Toolkit

Table 3: Essential Research Reagents & Resources

Item/Software Function/Application Example/Source
BEAST2 / RevBayes Primary software platforms for Bayesian phylogenetic analysis with integrated birth-death models. BEAST 2.7, RevBayes 1.2
FBD Model Package Implements the Fossilized Birth-Death process as a tree prior. BEAST2 package SA (Sampled Ancestors)
Birth-Death Skyline Package Enables estimation of time-varying birth and death rates. BEAST2 package bdsky
Gaussian Random Walk (GRW) Models time-varying parameters (e.g., Rₑ(t)) in a Bayesian framework. Available in BEAST2 and RevBayes model specification.
Tracer Diagnoses MCMC convergence and summarizes parameter posterior distributions. beast2.dev/tracer
TreeAnnotator Generates a maximum clade credibility tree from a posterior tree set. Distributed with BEAST2.
FigTree / IcyTree Visualizes time-calibrated phylogenetic trees, including node bars for uncertainty. github.com/revbayes/icyTree
Fossil Occurrence Database Sources for paleontological range data (e.g., occurrence ages). Paleobiology Database (paleobiodb.org)
Nextstrain / Augur Pipeline for real-time phylodynamic analysis of pathogen spread. nextstrain.org, docs.nextstrain.org

Conclusion

Bayesian birth-death analysis provides a powerful, statistically rigorous framework for reconstructing the dynamic histories of diversifying lineages, with profound implications for biomedical research. By moving from foundational concepts through practical implementation, troubleshooting, and comparative validation, researchers can confidently apply these models to critical questions in viral evolution, oncology, and immunology. The key takeaway is the method's unique ability to integrate prior knowledge, handle incomplete data, and quantify uncertainty in estimated rates and events—turning molecular sequences into a quantitative narrative of diversification. Future directions point toward integrating multimodal data (e.g., phenotypic traits, geographic data), developing multi-scale models linking cellular and population dynamics, and leveraging these inferences to predict future evolutionary trajectories for proactive drug and vaccine design. As computational power and methodological refinements advance, Bayesian birth-death models will become an indispensable tool in the translational research pipeline, directly informing strategies to combat evolving pathogens and therapeutic resistance.