QMix: An Efficient Program to Automatically Estimate Multi-Matrix Mixture Models for Amino Acid Substitution Process

Area of Science:

• Computational Biology
• Phylogenetics
• Evolutionary Biology

Background:

• Single-matrix amino acid (AA) substitution models struggle with site-specific rate heterogeneity.
• Multi-matrix mixture models address this limitation and offer improved phylogenetic accuracy.
• Estimating these complex models has been computationally challenging and lacked dedicated software.

Purpose of the Study:

• To develop a user-friendly computer program, QMix, for automated estimation of multi-matrix mixture models.
• To enhance the accuracy and speed of model parameter estimation using the QMaker algorithm.
• To provide a tool capable of handling large datasets and multi-threading for efficient computation.

Main Methods:

• Implementation of the QMix program based on LG4X and LG4M algorithms with enhancements.
• Utilized the QMaker algorithm for rapid and accurate parameter estimation.
• Employed multi-threading capabilities for processing thousands of genes and tested on 1471 HSSP alignments.

Main Results:

• Re-estimated mixture models (HP4X and HP4M) showed slight improvements over LG4X and LG4M in maximum likelihood tree construction.
• QMix efficiently estimated a four-matrix model from 200 HSSP alignments in approximately 10 hours on an 18-core computer.
• The program successfully handles varying numbers of matrices and large-scale datasets.

Conclusions:

• QMix provides an efficient and accessible solution for estimating complex multi-matrix amino acid substitution models.
• The developed models (HP4X, HP4M) offer enhanced phylogenetic inference capabilities.
• QMix facilitates advanced phylogenetic analyses by overcoming previous computational barriers.