Improving, validating, and interpreting amino acid substitution models

Aim 1 will improve the inference of amino acid substitution models by excluding likely alignment errors from the training data, and analyze the taxon-specific differences that we expect the removal of universal artifacts to better reveal. Aim 2 will train amino acid substitution models ondata specific to different types of protein structural characteristics, and test whether resulting mixture and/or partition models can improve phylogenetic inference. Aim 3 will estimate nonstationary amino acid models, validate them on against mean experimental effect sizes in deepmutational scanning data, and expand the method for application to gene trees rather than only ortholog sets.