Phylogenetic influence of complex, evolutionary models: a Bayesian approach

Molecular evolution recovers the history of living species by comparing genetic information, exploring genome structure and function from an evolutionary perspective. Here we infer substitution rates and ancestral reconstructions, to better understand mutation responses to some known biochemical phenomena. Mutation processes are commonly inferred using parsimony, maximum likelihood and Bayesian. Parsimony is not explicitly model-based, and is statistically biased due to unrealistic assumptions. The model-based maximum likelihood approaches become computationally inefficient while analyzing large or high-dimensional datasets, leaving little opportunities to incorporate complex evolutionary models. We implemented a posterior probability (Bayesian) approach that evaluates evolutionary models, applying it to primate mitochondrial genomes. The species nucleotide sequence data were augmented with ancestral states at the internal nodes of the phylogeny. We simplified probability calculations for substitution events along the branches by assuming that only up to one or two substitution events occurred per branch per site. These conditional pathway calculations introduce very little bias into the inferred reconstructions, while increasing the feasibility of incorporating complex evolutionary models with higher dimensions. Compositional bias tests, including functional predictions of ancestral tRNAs, show that ancestral sequences from the Bayesian approach are more biologically realistic than those reconstructed by maximum likelihood. To explore other model complexity, we allowed substitution rates to vary among sites by having a different model at each site. With a strand-symmetric model as the base model, asymmetric substitution probabilities for specific substitution types were varied among sites. This model would not be feasible with standard matrix exponentiation methods, particularly maximum likelihood. We observed for A--\u3eG and C--\u3eT substitutions almost linear, respectively, almost asymptotic responses (with some regional deviations). Note that the HMM models had no a priori response built in them. Observed responses fitted predictions from earlier gene by gene likelihood analyses. For A--\u3eG substitutions, deviations from the expected linear response correlated positively with the loop-forming propensity of the corresponding site in the mRNA secondary structure. In the COI region, C--\u3eT substitutions have a prominent dip, suggesting protection against mutations. The C--\u3eT substitution responses differed significantly between primate sub-groups defined based on their single genome A--\u3eG responses