A systematic evaluation revealed that detecting translated non-canonical ORFs from ribosome profiling data remains challenging

Non-canonical open reading frames (ORFs), which are ORFs that are not included in reference genome annotations, are gaining more and more research interest in recent years. While vast numbers of non-canonical ORFs have been identified with ribosome profiling (Ribo-Seq) by various state-of-the-art computational methods, the performance of these methods has not been assessed systematically. To this end, we evaluated the four most popular methods for translated non-canonical ORF prediction using various public datasets. We found that non-canonical ORFs predicted by different methods differ substantially and are not saturated at typical sequence depths. Furthermore, the precision and accuracy of all four methods are not satisfactory, especially for ORFs with near-cognate start codons. Based on these results, we suggest that improved sequence depth, biological repetitions, and translation initiation site profiling should be considered to obtain a high-quality catalog of translated non-canonical ORFs in future studies.