Fig. 2

Computational prediction and filtering of upstream-initiated open-reading frames. ORFs were predicted from AUG and non-AUG, near-cognate translation initiation sites in the 5′ leader of annotated protein-coding genes, and computationally extended to the first termination site encountered in the 5′ leader (upstream-terminated ORFs) or CDS (overlapping ORFs). Predicted ORFs were then screened through a series of heuristic filters including: 1) minimum RPF coverage in the 5′ leader, 2) minimum mRNA-seq coverage in CDS, 3) in-frame N-terminal extensions, 4) redundant isoforms, 5) minimum length with optimal RPF coverage, 6) sufficient SPECtre signal, and 7) removal of ambiguously annotated protein-coding transcripts.