mPartition: A Model-Based Method for Partitioning Alignments

Abstract

Maximum likelihood (ML) analysis of nucleotide or amino-acid alignments is widely used to infer evolutionary relationships among species. Computing the likelihood of a phylogenetic tree from such alignments is a complicated task because the evolutionary processes typically vary across sites. A number of studies have shown that partitioning alignments into sub-alignments of sites, where each sub-alignment is analyzed using a different model of evolution (e.g., GTR + I + G), is a sensible strategy. Current partitioning methods group sites into subsets based on the inferred rates of evolution at the sites. However, these do not provide sufficient information to adequately reflect the substitution processes of characters at the sites. Moreover, the site rate-based methods group all invariant sites into one subset, potentially resulting in wrong phylogenetic trees. In this study, we propose a partitioning method, called mPartition, that combines not only the evolutionary rates but also substitution models at sites to partition alignments. Analyses of different partitioning methods on both real and simulated datasets showed that mPartition was better than the other partitioning methods tested. Notably, mPartition overcame the pitfall of grouping all invariant sites into one subset. Using mPartition may lead to increased accuracy of ML-based phylogenetic inference, especially for multiple loci or whole genome datasets.