RiboWin: A computational framework to identify non-canonical open reading frames
Abstract Ribosome profiling is widely used to map translation across the genome, yet current sORF discovery pipelines are constrained by their reliance on annotated protein coding regions and by treating biological replicates primarily as tools for quantification rather than as strict filters for reproducibility. This results in two major issues, the systematic exclusion of pseudogene derived loci from analysis and the generation of extremely large candidate sets in which true translation events are difficult to distinguish from noise. To overcome these limitations, a five-stage framework was developed that expands the search space to include pseudogene loci while enforcing dual biological replicate concordance as a hard binary filter. When applied to four human cell lines with matched replicates and benchmarked against existing tools, the framework produces dramatically smaller and more reliable candidate sets, reducing thousands to millions of predicted sORFs down to a small number of high confidence concordant loci. Across HeLa cell cycle conditions, a limited set of unique concordant sORFs was identified spanning multiple transcript classes including upstream ORFs, retained introns, and nonsense mediated decay transcripts, along with a pseudogene derived candidate. Notably, the pseudogene encoded sORF RPL26P19 was detected consistently in HeLa cells but missed or inconsistently identified by existing tools, while a separate pseudogene sORF from UBE3AP2 was observed specifically in normal primary cell types and absent in HeLa. Both candidates were later found to be supported by entries in OpenProt, even though this was not known during the analysis, which gives added confidence that the results are genuine and not biased by prior expectations. Overall, these findings suggest that pseudogenes are not just inactive remnants in the genome but can act as real, functional sources of translated sORFs in a cell type specific manner. At the same time, it becomes clear that most current tools are not designed to detect these signals effectively. This highlights the need for approaches that both broaden the regions being searched and apply stricter reproducibility checks, so that truly meaningful translation events can be identified more reliably.