All Together Now: Joint Codon Alignment, Ohmic Networks, and Error-Controlled Selection Inference in HyPhy 3

Read the full article See related articles

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Codon-based maximum likelihood models are the workhorse of molecular selection analysis, yet standard workflows assume that the input sequence alignment and phylogenetic tree are known without error. In practice, heuristic aligners ignore the genetic code and phylogeny, which packs non-homologous codons into shared columns across insertion-deletion loops and inflates gene-wide false-positive rates above 80%. This sequential division of labor is statistically suboptimal and computationally wasteful. Here, we present WALRUS ( W heatstone A lignment and L ikelihood R ate U nification for S election), a unified engine in H y P hy 3 ( https://github.com/veg/hyphy-3 ) that maps unaligned coding sequences to a coordinate-consistent selection atlas in one run. From a single Inside–Outside likelihood table, the pipeline repairs reading frames, co-estimates a codon alignment and Ohmic reticulate network, models directional base-composition drift, and evaluates eight selection tests (BUSTED-E, aBSREL, RELAX, FEL, dFEL, MEME, Contrast-FEL, and PRIME). Across 10,020 simulated alignments, WALRUS recovers 94.6% of true columns on the 800 INDELible replicates and cuts indel-driven false positives from 44.7%– 58.1% to 4.4% under BUSTED (21.1%– 21.4% to 1.4% under BUSTED-E). Under Zipfian indel saturation, false positives drop four-fold (33.3%– 83.3% to 17.0%– 20.5%), while 11, 190 avian orthogroup runs execute 6.6–35.8-fold faster and cut housekeeping false positives 13-fold (28-fold on ultraconserved genes) proteome-wide.

Article activity feed