SQANTI-evidence: isoform-resolved, evidence-tiered structural annotation of unannotated genomes from long-read RNA-seq

Long-read RNA sequencing captures full-length transcripts, yet current annotation pipelines fail to properly curate the data for gene prediction. We present SQANTI-evidence, a workflow that annotates a genome from long reads alone, without a reference annotation or protein database. SQANTI3-curated transcripts enter the annotation as observed, with isoforms and untranslated regions; AUGUSTUS predicts genes at loci without read support. In benchmarks, the curated layer was the most precise gene set in five of seven species, recovered twice as many exact isoforms as BRAKER in mammals and placed 3' ends nearest the poly(A) site. We annotated 13 ERGA datasets.