Bootstrap Analysis: Measuring Statistical Support in Phylogenetics

When you build a phylogenetic tree, how confident can you be that the relationships it shows are correct? Bootstrap analysis is the most widely used method for assessing statistical support in phylogenetics, providing a measure of how robust each clade is to sampling variation in your data.

What is Bootstrap Analysis?

Bootstrap analysis, introduced to phylogenetics by Joseph Felsenstein in 1985, uses resampling with replacement to assess the stability of phylogenetic relationships. The basic idea is simple: if a clade is well-supported by the data, it should appear consistently even when we randomly perturb the dataset.

The Bootstrap Procedure

Create pseudoreplicates: Generate new datasets by randomly sampling columns (characters/sites) from your alignment with replacement
Build trees: Construct a phylogenetic tree from each pseudoreplicate using your chosen method
Count clades: For each clade in your original tree, count how often it appears in the bootstrap trees
Calculate support: The percentage of bootstrap trees containing a clade is its bootstrap support value

Original alignment (5 characters):
  A B C D E
  | | | | |
Taxon1: A T G C T
Taxon2: A T G C A
Taxon3: G C A T T
Taxon4: G C A T A

Bootstrap replicate 1 (sample with replacement):
  B B D E A  (columns 2,2,4,5,1)

Bootstrap replicate 2:
  C A E E D  (columns 3,1,5,5,4)

... repeat 100-1000 times

Interpreting Bootstrap Values

Bootstrap values range from 0 to 100 (or 0 to 1 as proportions). But what do these numbers actually mean?

Bootstrap Value	Interpretation	Common Usage
≥95%	Very strong support	Highly confident in clade
90-94%	Strong support	Generally reliable
70-89%	Moderate support	Tentatively supported
50-69%	Weak support	Uncertain
<50%	No support	Should not be considered reliable

The 70% Threshold

Hillis and Bull (1993) showed that bootstrap values ≥70% correspond to ≥95% probability that a clade is real under certain conditions. However, this relationship depends on many factors, so treat these thresholds as guidelines rather than hard rules.

How Many Bootstrap Replicates?

The number of bootstrap replicates affects the precision of your support estimates:

100 replicates: Minimum for publication; standard error ~5%
500 replicates: Good precision; standard error ~2%
1000 replicates: High precision; standard error ~1.5%
10000+ replicates: For precise estimates of low support values

For most purposes, 100-500 replicates are sufficient. If a clade has 73% support with 100 replicates, it won't change dramatically with 1000 replicates.

Bootstrap vs. Posterior Probability

Bayesian analysis provides posterior probabilities instead of bootstrap values. These two measures are NOT directly comparable:

Aspect	Bootstrap	Posterior Probability
Interpretation	Frequency of clade in resampled data	Probability clade is correct given data & model
Typical "strong" threshold	≥70-75%	≥0.95
Tendency	Often conservative	Can be overconfident with poor models

Don't Compare Numbers Directly

A bootstrap of 70% and a posterior probability of 0.95 may indicate similar actual support. Don't assume PP 0.70 = Bootstrap 70% - they measure different things.

Limitations of Bootstrap

What Bootstrap Doesn't Tell You

Not probability: A 95% bootstrap doesn't mean 95% probability the clade is correct
Only assesses sampling variance: Doesn't account for model misspecification
Assumes independent characters: Linked sites violate this assumption
Sensitive to data quality: Errors in alignment propagate through analysis

When Bootstrap Can Mislead

Long-branch attraction: High support for incorrect groupings
Compositional bias: Similar nucleotide frequencies causing spurious groupings
Gene tree/species tree conflict: High support for gene tree that differs from species tree

Alternatives to Standard Bootstrap

Ultrafast Bootstrap (UFBoot)

IQ-TREE's ultrafast bootstrap is ~10-40x faster than standard bootstrap while maintaining accuracy. It uses a different resampling strategy optimized for ML trees.

SH-aLRT

The Shimodaira-Hasegawa approximate likelihood ratio test provides fast support assessment. Often used alongside UFBoot for corroboration.

Transfer Bootstrap Expectation (TBE)

TBE measures how often taxa appear on the same side of a branch, providing more stable estimates for large trees.

Best Practices

Use enough replicates: At least 100, preferably 500+
Report the method: Specify standard vs. ultrafast bootstrap
Don't over-interpret: High bootstrap doesn't guarantee correctness
Use multiple measures: Combine bootstrap with other support measures
Consider data quality: Bootstrap can't fix bad alignments

Run Bootstrap Analysis

Calculate bootstrap support for your phylogenetic trees directly in PhyloVerse. Visualize support values on branches and export publication-ready figures.

Launch PhyloVerse