How many bootstrap replicates are necessary?

Area of Science:

• Computational Biology
• Phylogenetics
• Bioinformatics

Background:

• Phylogenetic bootstrapping (BS) is crucial for assessing confidence in phylogenetic trees, particularly with maximum likelihood (ML) inference.
• Traditional BS methods are computationally intensive, limiting the number of replicates (typically 100) and thus the accuracy of support values.
• A recent faster BS algorithm enables large-scale experimental studies on the impact of replicate number.

Purpose of the Study:

• To propose and evaluate novel stopping criteria for phylogenetic bootstrapping (BS) to determine optimal replicate numbers.
• To experimentally assess the effect of the number of BS replicates on the quality and accuracy of support values in ML phylogenetic inference.
• To make robust BS computationally practical for a wider range of datasets.

Main Methods:

• Implemented and tested novel runtime stopping criteria for phylogenetic bootstrapping (BS).
• Conducted a large-scale experimental study using 17 diverse real-world DNA datasets (single- and multi-gene, 125-2,554 taxa).
• Compared support values generated with proposed stopping criteria against reference values from best ML trees.

Main Results:

• Proposed stopping criteria typically halt computations between 100-500 replicates, with some conservative criteria extending to several thousand.
• Generated support values correlate at >99.5% with reference values, demonstrating high accuracy.
• The number of recommended replicates varies significantly across datasets, even those of comparable size.

Conclusions:

• The study provides the first experimental assessment of replicate number effects on BS support value quality.
• The proposed stopping criteria are validated, enabling practitioners to obtain robust support values without guesswork.
• Efficient BS with ML inference is now computationally feasible for most datasets, with typical replicate counts between 100-500.