Statistical Inconsistency of Error-correction Objectives for Perfect Phylogenies

Area of Science:

• Computational Biology
• Evolutionary Biology
• Phylogenetics

Background:

• The binary perfect phylogeny model assumes mutations arise once and are never lost.
• Phylogeny inference often minimizes error corrections (flips) to fit observed data to a perfect phylogeny.

Purpose of the Study:

• To test the assumption that minimizing error corrections leads to the true evolutionary tree.
• To investigate the statistical consistency of error-correction objectives in phylogeny inference.

Main Methods:

• Developed a generative model with independent errors to analyze phylogeny inference.
• Proved statistical inconsistency of error-correction objectives for all positive error rates.
• Conducted simulations using error rates from single-cell sequencing data.

Main Results:

• Error-correction objectives are statistically inconsistent; the minimum-cost tree may not be the true tree.
• Topological features, particularly deeper and imbalanced trees, are favored over balanced ones at typical error rates.
• Incorrect trees are preferred over true trees in over 50% of simulated cases with increasing tree size.

Conclusions:

• Minimizing implied errors does not guarantee accurate phylogeny inference due to systematic bias.
• Inconsistency is a practical issue, not just a theoretical edge case, particularly in single-cell sequencing data.
• Understanding topological bias is crucial for interpreting phylogenetic results.