Efficient inference, potential, and limitations of site-specific substitution models

Area of Science:

• Evolutionary biology
• Computational biology
• Phylogenetics

Background:

• Natural selection creates varied evolutionary patterns across genomes, with some sites conserved and others dynamic.
• Phylogenetic models often simplify these patterns by using discrete site classes, which may not capture full complexity.
• Accurate phylogenetic reconstruction requires accounting for complex sequence evolution constraints.

Purpose of the Study:

• To develop an efficient algorithm for estimating complex phylogenetic models with site-specific evolutionary preferences.
• To assess the accuracy of estimating these complex models from simulated and real-world data.
• To explore the implications of site-specific preferences and divergence saturation on inferring evolutionary history.

Main Methods:

• Developed an iterative approximate maximum likelihood algorithm for estimating site-specific preferences.
• Used simulated data to test the accuracy of the algorithm in estimating complex models.
• Applied the algorithm to large HIV pol alignments to estimate site-specific preferences and divergence patterns.

Main Results:

• The algorithm accurately estimates site-specific preferences from large datasets with moderately diverged sequences.
• Joint estimation of site-specific rates, preferences, and branch lengths can present identifiability challenges.
• Ignoring site variation leads to underestimation of phylogenetic branch lengths.
• Estimated site-specific preferences from HIV pol alignments correlate with intra-host fitness costs.
• Analysis indicates divergence saturation in HIV evolution after a few hundred years.

Conclusions:

• The developed algorithm provides an efficient way to estimate complex, site-specific evolutionary models.
• Site-specific evolutionary preferences are crucial for accurate phylogenetic reconstruction.
• Divergence saturation in HIV evolution may explain difficulties in inferring deep divergence times using molecular clock methods.