Modeling multi-species RNA modification through multi-task curriculum learning

Abstract N6-methyladenosine (m6A) is the most pervasive modification in eukaryotic mRNAs. Numerous biological processes are regulated by this critical post-transcriptional mark, such as gene expression, RNA stability, structure and translation. Recently, various experimental techniques computational methods have been developed to characterize transcriptome-wide landscapes of m6A for understanding its underlying mechanisms functions mRNA regulation. However, generally costly time-consuming, while existing models usually designed only site prediction a single-species significant limitations accuracy, interpretability generalizability. Here, we propose highly interpretable framework, called MASS, based on multi-task curriculum learning strategy capture features across multiple species simultaneously. Extensive experiments demonstrate superior performances MASS when compared state-of-the-art methods. Furthermore, contextual sequence captured can be explained known binding motifs related RNA-binding proteins, which also help elucidate similarity difference among species. In addition, predicted profiles, further delineate relationships between properties regulation, including translation, histone modification. summary, may serve useful tool characterizing studying regulatory code. The source code downloaded from https://github.com/mlcb-thu/MASS.