Phylogenetic Bootstrap Interpretation
The phylogenetic bootstrap is a resampling technique that estimates the support for each branch in a tree. It is not a measure of accuracy or probability that the branch is true. This guide is for molecular biologists, bioinformaticians, and graduate students who need a practical, source bounded framework for interpreting bootstrap values in their phylogenomic or phylogenetic analyses. NCBI Bookshelf provides foundational references on this method.
You will learn about core concepts, decision criteria, a step by step workflow, common mistakes, and the limits of bootstrap interpretation. EMBL EBI Training offers structured lessons on phylogenetics that complement this material. By the end of this guide you should be able to read a bootstrap value with confidence and avoid typical overinterpretation errors.
At a Glance
| Concept | Interpretation | Typical Thresholds |
|---|---|---|
| Bootstrap Proportion (BP) | Proportion of resampled datasets that recovered a given branch. | >70% moderate support, >95% strong support. |
| SH aLRT | Shimodaira Hasegawa like approximate likelihood ratio test. | >80% strong support. |
| UFboot | Ultrafast bootstrap approximation for maximum likelihood. | >95% high support. |
| Transfer Bootstrap Expectation (TBE) | Expectation of transfer distance between bootstrap trees. | >90% strong support. |
| Nonparametric bootstrap | Resample alignment columns with replacement. | Widely used but sensitive to model misspecification. |
Use these thresholds as guidelines, not absolute cutoffs. The biological context and data quality always override any arbitrary number.
Core Concepts and Interpretation Framework
The bootstrap works by generating many pseudoreplicates of your alignment. You resample columns with replacement to create a new alignment of the same length. You then build a tree for each pseudoreplicate. The bootstrap value for a branch is the fraction of pseudoreplicate trees that contain that branch. A value of 95 means that 95 out of 100 pseudoreplicates recovered that same branch.
A high bootstrap value indicates that the branch is consistently recovered despite random variation in the alignment. A low value suggests that the branch is sensitive to which characters are sampled. Galaxy Training Network provides hands on tutorials that show how to run bootstrap analyses in standard phylogenetic workflows.
The key point is that a 70% bootstrap value does not mean there is a 70% chance the branch is correct. Bootstrap measures the stability of the tree under resampling, not the probability of the true tree. This distinction is crucial for interpretation. When you see a branch with 50% support, it means that branch appeared in only half of the replicates. That branch should be considered poorly supported.
Decision Criteria for Assessing Bootstrap Support
You need to decide which bootstrap threshold to apply. This depends on the number of characters, the evolutionary rate heterogeneity, and the method used. For small alignments (fewer than 500 sites), bootstrap values tend to be low even for correct branches. In that case, you may accept values above 60% as moderate support. For large phylogenomic datasets (thousands of loci), branches are often highly supported, and any value below 90% may indicate conflict or model inadequacy.
Check the model fit before interpreting bootstrap values. A poor evolutionary model can inflate or deflate support. Bioconductor offers packages like phangorn and ape that include model testing and bootstrap implementation. Use a model selection tool (e.g., ModelTest or ModelFinder) and confirm that the model fits the data.
Also consider the taxon sampling. Sparse taxonomic sampling can lead to long branches that appear highly supported but are artifacts. Bootstrap values on long branches require extra scrutiny. When you see a terminal branch with 100% support and no close relatives sampled, question it. Incomplete lineage sorting or recombination can also cause high bootstrap support for incorrect groupings.
Practical Workflow for Bootstrap Analysis
Follow these steps to generate and interpret bootstrap values in a reproducible manner.
- Prepare your multiple sequence alignment. Remove ambiguous regions. Use a tool like MAFFT or Muscle and then trim with Gblocks or TrimAl.
- Select the best fit substitution model. Use ModelFinder (integrated in IQ TREE) or jModelTest. This step is mandatory.
- Choose your tree inference method. Maximum likelihood with IQ TREE or RAxML is standard. For neighbor joining, bootstrap is simpler but less robust.
- Run the bootstrap. For maximum likelihood, use 1000 replicates as a minimum. For ultrafast bootstrap (UFboot), 1000 replicates are sufficient and fast.
- After analysis, examine the tree with bootstrap values. Collapse branches with support below your chosen threshold for a conservative tree.
- Assess convergence. Check that the bootstrap values stabilize after a certain number of replicates. You can use a plot of cumulative support versus replicate number.
NCBI Sequence Read Archive may contain raw reads used to generate the alignment. Always document the source data. For example, Acquired HIV 1 Drug Resistance and Molecular Transmission Networks in Zhongwei, Ningxia, China used phylogenetic bootstrap to assess transmission clusters. Their bootstrap support values above 95% were considered robust.
Common Mistakes in Bootstrap Interpretation
The most frequent mistake is treating bootstrap values as probabilities of truth. A 70% bootstrap does not mean the branch has a 70% chance of being correct. It means that 70% of pseudoreplicates recovered that branch. The true tree could still be different.
Another error is applying a single threshold across all analyses. A 70% threshold may be too strict for small datasets and too lenient for large ones. Use the literature from your field as a guide. For viral phylogenies, often 70% is considered moderate. For genomic studies, 100% is expected for many branches.
Ignoring alignment quality is another mistake. Gappy or misaligned regions produce artificially high bootstrap values because few informative sites remain. Always trim alignments before running bootstraps.
Some researchers report bootstrap values without specifying the method (e.g., nonparametric vs. UFboot). Always state the method. UFboot can give slightly higher values than standard bootstrap for the same data.
Finally, do not use bootstrap values as the sole criterion for evolutionary conclusions. Combine them with alternative support values like SH aLRT or posterior probabilities from Bayesian analysis. Are interphylum spiralian relationships resolvable? illustrates how different support measures can conflict, especially in deep phylogenetic nodes.
Limits and Uncertainty of Bootstrap Values
Bootstrap values have inherent limitations. They do not account for systematic error such as long branch attraction, compositional bias, or model misspecification. If your model is wrong, the bootstrap may consistently support an incorrect tree. This is known as high bootstrap support for the wrong tree.
Bootstrap values also depend on the number of characters. An alignment with 1000 sites will yield higher bootstrap values than one with 100 sites, even if the underlying signal is the same. This means you cannot compare bootstrap values across studies with different alignment lengths without caution.
Another limit is that bootstrap values are not independent across branches. A branch with 100% support may be artificially high if neighboring branches are poorly supported. The resampling process is linked.
Consider using gene wise bootstrapping for phylogenomic datasets. Instead of bootstrapping sites, you can bootstrap loci. This accounts for genealogical variation across the genome. A comprehensive phylogenomic framework for cycads (Cycadales) used gene bootstrap to assess congruence across loci.
Always report the bootstrap method, number of replicates, and the software version. This allows others to reproduce your results. Characterization of the Complete Mitochondrial Genome of Nibea chui provides an example of transparent reporting of bootstrap settings.
Frequently Asked Questions
1. What does a bootstrap value of 70% mean?
It means that in 70 out of 100 pseudoreplicates, that particular branch appeared. It does not mean the branch has a 70% chance of being correct. It indicates moderate support. The interpretation should consider the size of the dataset and the method used.
2. Can I get 100% bootstrap support for an incorrect branch?
Yes. This can happen when the evolutionary model is misspecified, when there is long branch attraction, or when the alignment contains systematic bias. High bootstrap support does not guarantee the branch is biologically valid. Always test alternative models.
3. Should I bootstrap every phylogenetic analysis?
Yes, if you want to communicate confidence in your tree. Exceptions include exploratory analyses or very small alignments where bootstrap is not informative. For published studies, bootstrap values (or equivalent measures) are expected.
4. What is the difference between bootstrap proportion and SH aLRT?
Bootstrap proportion is based on resampling the alignment. SH aLRT is a likelihood based test that compares alternative tree topologies. SH aLRT values are generally higher than bootstrap values for the same branch. Some researchers report both to provide complementary support.
References and Further Reading
- NCBI Bookshelf , Contains authoritative chapters on phylogenetic inference and bootstrap methodology.
- EMBL EBI Training , Offers online courses on phylogenetics and bootstrapping.
- Galaxy Training Network , Practical tutorials on running phylogenetic analyses with bootstrap.
- Bioconductor , Software packages (ape, phangorn) for bootstrap and tree evaluation.
- Acquired HIV 1 Drug Resistance and Molecular Transmission Networks in Zhongwei, Ningxia, China , Applied example of bootstrap interpretation in viral phylogenetics.
- Are interphylum spiralian relationships resolvable? , Discusses conflicts between support measures in deep phylogeny.
- A comprehensive phylogenomic framework for cycads (Cycadales) , Uses gene bootstrap for phylogenomic support.
- Characterization of the Complete Mitochondrial Genome of Nibea chui , Example of reporting bootstrap values in mitogenome analysis.
- Malaria Vector Surveillance in Indonesia , Demonstrates bootstrap use in species identification.
- HIV 1 Resistance to Dolutegravir in Benin , Clinical context where bootstrap informed transmission studies.