Illuminating phage dark matter beyond the reach of structural alignments with PhageFlare

PhageFlare predicts functions for phage proteins without significant Foldseek structural matches, reaching macro-averaged F1 scores of 0.871, 0.889, and 0.738 across three structure-disjoint benchmarks.

• Why it matters: Many phage proteins remain uncharacterized because structural-similarity annotation fails when no significant structural match exists, limiting discovery in viral dark matter.
• What they did: The deep learning framework was evaluated on three structure-disjoint benchmarks using Foldseek E-value thresholds of 10-3, 10-4, and 10-2, with protein language model selection tested (ESM-C vs ESM-2 and ProtBert).
• The result: PhageFlare annotated 38.0% and 51.8% of Foldseek-unresolved proteins in the efam and Metagenomic Gut Virus datasets, and identified lysogenic potential in 1,866 prophage-designated genomes misclassified as lytic by BACPHLIP.