Discovering misannotated lncRNAs using deep learning training dynamics

Abstract

Motivation Recent experimental evidence has shown that some long noncoding RNAs (lncRNAs) contain small open reading frames (sORFs) that are translated into functional micropeptides, suggesting that these lncRNAs are misannotated as noncoding. Current methods to detect misannotated lncRNAs rely on ribosome-profiling (Ribo-Seq) and mass-spectrometry experiments, which are cell-type dependent and expensive. Results Here, we propose a computational method to identify possible misannotated lncRNAs from sequence information alone. Our approach first builds deep learning models to discriminate coding and non-coding transcripts and leverages these models’ training dynamics to identify misannotated lncRNAs—i.e. lncRNAs with coding potential. The set of misannotated lncRNAs we identified significantly overlap with experimentally validated ones and closely resemble coding protein sequences as evidenced by significant BLAST hits. Our analysis on a subset of misannotated lncRNA candidates also shows that some ORFs they contain yield high confidence folded structures as predicted by AlphaFold2. This methodology offers promising potential for assisting experimental efforts in characterizing the hidden proteome encoded by misannotated lncRNAs and for curating better datasets for building coding potential predictors. Availability Source code is available at https://github.com/nabiafshan/DetectingMisannotatedLncRNAs. Supplementary information Supplementary data are available at Bioinformatics online.

Shining a light on the dark proteome: Non‐canonical open reading frames and their encoded miniproteins as a new frontier in cancer biology

“…To deliver on this promise, the set of computational and experimental tools to find and functionally validate ncORFs and their miniprotein products requires ongoing expansion and improvement. While current identification of ncORFs mainly relies on Ribo‐seq and mass spectrometry, new computational methods (Nabi et al, 2023 ) potentiate higher‐confidence and higher‐throughput discovery. For instance, the TIS Transformer program, which uses artificial intelligence to map the human genome, was able to detect ncORFs and predict those that encode miniproteins with high performance (Clauwaert et al, 2023 ).…”

Micropeptides: potential treatment strategies for cancer

“…It does not require a prelabelled dataset consisting of positively identified but misannotated lncRNAs. In contrast, it is based on Ribo-Seq data, which are used to discern lncRNAs that have been previously misannotated [ 52 ]. Of course, although bioinformatics approaches are essential for identifying sORFs, relying on only one technique to predict the coding potential of sORFs in a specific study is insufficient, and other experimental methods should be combined as appropriate to confirm whether these sORFs can be translated into small functional proteins.…”

Discovering microproteins: making the most of ribosome profiling data

“…Generation of a reference set incorporated along with the known ORFs would allow researchers to use publicly available RNAseq data to infer which smORFs are differentially regulated in a given disease or perturbation of interest without having to perform Ribo-seq and call smORFs in every new system of study. For those interested to investigate lncRNA coding potential, several sequence- and evolutionary conservation-based tools [ 90 , 91 ], and more recently deep learning models, have been developed [ 92 , 93 ] to identify cryptic ORFs using in silico prediction. As has recently become evident, lncRNAs tend to encode young proteins [ 19 ] and thus traditional ORF prediction methods which rely on length-biased and evolutionary conservation-biased methods would not be able to discern coding potential efficiently [ 46 ].…”