Developing codon model tools to improve detection of natural selection

Codon-based evolutionary models are frequently employed to infer natural selection from protein-coding DNA. However, most codon models assume synonymous substitutions are a suitable reference for the neutral background substitution rate—an assumption increasingly violated by selection on genomic codon usage, mRNA-level constraints, and other forces. Such confounders can bias key model parameters and inflate evidence for positive selection. Here we introduce Multiclass Synonymous Substitution (MSS) models, which allow multiple synonymous rate classes up to a codon-pair–level reversible framework. We demonstrate that MSS models are a substantially better fit to empirical data than traditional codon models, and show that empirical synonymous rates are highly heterogeneous. We then integrate MSS into an episodic positive selection test (BUSTED+S+MSS) and demonstrate that MSS models enable more robust detection of adaptive evolution when synonymous codon usage is constrained across the genome. Beyond improving codon models’ fidelity with MSS, there is also a pressing need for a generalist computational workflow that streamlines exploratory selection analysis on large sequencing datasets, particularly datasets of pathogen genomes. To that end, we present CAPHEINE, a reproducible high-throughput workflow that processes input sequence data, performs a range of selection tests, and summarizes the output in a clear and concise manner.