🌐 English|한국어

Analysis Guide

This guide provides detailed workflows for running phylogenetic analyses in PhyloForester.

Overview

PhyloForester supports three main approaches to phylogenetic tree reconstruction:

  1. Parsimony: Finds trees minimizing character state changes

  2. Maximum Likelihood: Finds trees maximizing probability of observed data

  3. Bayesian Inference: Estimates posterior probability distribution of trees

Each method has strengths and is suited for different datasets and research questions.

Choosing an Analysis Method

Parsimony

Best for:

  • Morphological data

  • Small to medium datasets (<100 taxa)

  • Discrete character states

  • Pedagogical purposes

Advantages:

  • Fast computation

  • No model assumptions

  • Easy to interpret

Limitations:

  • Assumes equal rates across branches

  • Can be inconsistent with long branches

  • No statistical framework for support

Maximum Likelihood

Best for:

  • Molecular data (DNA/protein sequences)

  • Large datasets (100+ taxa)

  • Model-based inference

Advantages:

  • Statistical framework

  • Model flexibility

  • Bootstrap support values

Limitations:

  • Computationally intensive

  • Requires model selection

  • Less suited for morphology

Bayesian Inference

Best for:

  • Complex evolutionary models

  • Integrating prior information

  • Uncertainty quantification

Advantages:

  • Full probabilistic framework

  • Posterior probabilities

  • Handles complex models well

Limitations:

  • Very computationally intensive

  • Convergence assessment required

  • Prior specification needed

Parsimony Workflow

Step 1: Prepare Data

Ensure your datamatrix:

  • Has clear character definitions

  • Minimal missing data (?)

  • Proper inapplicable coding (-)

Step 2: Create Parsimony Analysis

  1. Right-click datamatrix → New Analysis → Parsimony

  2. Configure parameters:

    Replicates: 100
Hold: 1000
TBR: Yes (tree bisection-reconnection)
Mult: 10 (random addition sequences)

Step 3: Run Analysis

  1. Click Start Analysis

  2. TNT performs heuristic search

  3. Progress shown as percentage

  4. Typical runtime: seconds to minutes

Step 4: Examine Results

Check the Log tab for:

  • Number of trees found

  • Tree length (total character changes)

  • Consistency Index (CI)

  • Retention Index (RI)

Trees Tab:

  • Strict consensus tree shown

  • Bootstrap values (if requested)

  • Branch lengths (steps)

Maximum Likelihood Workflow

Step 1: Prepare Data

For DNA sequences:

  • Aligned sequences required

  • IUPAC ambiguity codes supported

  • Gap coding as missing (?) or 5th state

Step 2: Model Selection

IQTree can auto-detect best model:

  1. Enable Auto-detect model

  2. IQTree tests all standard models

  3. Best model selected by AIC/BIC

Or specify model manually:

  • JC69: Jukes-Cantor (equal rates)

  • K80/K2P: Kimura 2-parameter (transitions/transversions)

  • HKY: Has​egawa-Kishino-Yano

  • GTR: General time reversible (most parameters)

Step 3: Create ML Analysis

  1. Right-click datamatrix → New Analysis → Maximum Likelihood

  2. Configure:

    Model: Auto-detect
Bootstrap: 1000
Algorithm: Standard

Step 4: Run Analysis

  1. Click Start Analysis

  2. Model testing phase (if auto-detect)

  3. Tree search phase

  4. Bootstrap phase (if enabled)

  5. Typical runtime: minutes to hours

Step 5: Interpret Results

Best Tree:

  • Maximum likelihood tree topology

  • Branch lengths (substitutions/site)

  • Log-likelihood score

Bootstrap Support:

  • Values 0-100 on nodes

  • ≥70 generally considered significant

  • ≥95 strong support

Bayesian Workflow

Step 1: Prepare Data

Similar to ML, but Bayesian is more flexible:

  • Can handle complex partitions

  • Morphology + molecules combined

  • Clock models for dating

Step 2: Set Priors

Substitution Model:

  • Often use GTR+Γ for DNA

  • Mk model for morphology

Tree Prior:

  • Uniform (default)

  • Birth-death process

  • Yule model

Branch Length Prior:

  • Exponential distribution

  • Compound Dirichlet

Step 3: Configure MCMC

  1. Right-click datamatrix → New Analysis → Bayesian

  2. Set parameters:

    Generations: 1,000,000
Sample frequency: 1000
Burnin: 0.25 (25%)
Chains: 4 (2 heated)

Short run for testing:

Generations: 100,000
Sample: 100
Burnin: 0.25

Standard run:

Generations: 10,000,000
Sample: 1000
Burnin: 0.25

Step 4: Run Analysis

  1. Click Start Analysis

  2. MrBayes runs MCMC chains

  3. Monitor: - Average standard deviation of split frequencies (should approach 0.01) - Potential Scale Reduction Factor (should approach 1.0)

  4. Typical runtime: hours to days

Step 5: Assess Convergence

Check the Log for:

  • ASDSF < 0.01: Chains converged

  • ESS > 200: Sufficient sampling

  • Stable log-likelihood traces

If not converged:

  • Run more generations

  • Increase sample frequency

  • Simplify model

Step 6: Examine Posterior

Consensus Tree:

  • 50% majority rule consensus

  • Posterior probabilities on nodes

  • ≥0.95 generally considered strong support

Credible Sets:

  • 95% credible set of trees

  • Topology uncertainty quantified

Character Mapping

After obtaining trees, map characters to visualize evolution.

Fitch Parsimony Mapping

PhyloForester uses Fitch’s algorithm for ancestral state reconstruction.

  1. Open analysis with trees

  2. Select a tree in Trees tab

  3. Click Map Character

  4. Select character from list

  5. Tree shows: - Ancestral states at nodes - State changes on branches (synapomorphies) - Colored by state

Interpreting Mapped Trees

  • Node labels: Reconstructed ancestral states

  • Branch annotations: Character changes

  • Colors: Different states

  • Ambiguous: Multiple optimal reconstructions shown

Use cases:

  • Identify evolutionary transitions

  • Locate homoplasy (parallel/convergent evolution)

  • Support morphological hypotheses

Comparing Analyses

It’s valuable to compare results across methods.

Topology Comparison

  1. Run multiple analysis types on same datamatrix

  2. Compare tree topologies visually

  3. Note areas of agreement/disagreement

Key questions:

  • Do methods agree on major clades?

  • Where do topologies differ?

  • Are differences in weakly supported regions?

Support Value Comparison

  • Parsimony: Bootstrap (if run)

  • ML: Bootstrap percentages

  • Bayesian: Posterior probabilities

Generally:

  • Bayesian PP ≥ 0.95 ≈ ML bootstrap ≥ 70%

  • Bayesian tends to give higher values

  • ML bootstrap more conservative

Troubleshooting Analyses

Analysis Won’t Start

Check:

  1. External software path set correctly (Preferences)

  2. Software executable has permissions

  3. Datamatrix not empty

  4. No special characters in names

Analysis Fails Immediately

Check:

  1. Log tab for error messages

  2. Datamatrix format correct

  3. Missing data not excessive

  4. Character definitions valid

Analysis Runs Forever

For Bayesian:

  • May take days - check convergence diagnostics

  • Consider reducing generations for testing

For ML:

  • Large datasets take time

  • Consider reducing bootstrap replicates temporarily

For Parsimony:

  • Usually fast; if slow, reduce Hold parameter

Poor Support Values

Common reasons:

  • Insufficient data

  • Conflicting signal

  • Model misspecification

  • Need more bootstrap replicates

Solutions:

  • Add more characters/taxa

  • Try different models

  • Increase replicates

  • Partition data

Best Practices

Data Preparation

  1. Carefully define characters

  2. Minimize missing data

  3. Check for typos in taxon names

  4. Validate alignment (for sequences)

Parameter Selection

  1. Start with default/recommended values

  2. Do quick test runs first

  3. Increase rigor for final analyses

  4. Document all parameters used

Quality Control

  1. Always check log files

  2. Verify convergence (Bayesian)

  3. Compare multiple runs

  4. Examine support values critically

Publication

When publishing, report:

  • Software versions

  • All parameter settings

  • Run statistics (length, likelihood, etc.)

  • Support measures

  • Convergence diagnostics (Bayesian)

Next Steps