PhyloScore: Improved covariation score for RNA alignments using phylogeny

Abstract

Motivation: Structured RNAs can carry out complex biological functions similar to proteins, but our understanding of RNA structure and function lags far behind that of proteins. Covariation signals arising from RNA evolution can be recovered from sequence alignments and provide valuable insights into structure and function. However, such signals can also arise purely as a result of phylogenetic relatedness. We aimed to develop a method which can make use of the phylogenetic tree to distinguish better between real and false signal. Results: PhyloScore predicts the manually curated secondary structure annotations in Rfam more accurately than the best currently used scores.On 48 Rfam families with experimentally determined structures, PhyloScore increases the mean average precision for recovering the annotated base pairs from 0.679 (R-scape's APC-corrected G-test) to 0.714 under cross-validation, and improves 41 of the 48 families. Its p-values are discriminative across families, making them valuable for testing whether a non-coding RNA has a conserved structure. Availability: PhyloScore is available as a pip package "phylo score" under MIT Licence. The code repository is at https://github.com/ soedinglab/phylo_score. The version described in this paper, including test data, is archived at https://doi.org/10.5281/zenodo.23401723