Deep analysis of RNA N⁶-adenosine methylation (m⁶A) patterns in human cells

N⁶-adenosine methylation (m⁶A) is the most abundant internal RNA modification in eukaryotes, and affects RNA metabolism and non-coding RNA function. Previous studies suggest that m⁶A modifications in mammals occur on the consensus sequence DRACH (D = A/G/U, R = A/G, H = A/C/U). However, only about 10% of such adenosines can be m⁶A-methylated, and the underlying sequence determinants are still unclear. Notably, the regulation of m⁶A modifications can be cell-type-specific. In this study, we have developed a deep learning model, called TDm6A, to predict RNA m⁶A modifications in human cells. For cell types with limited availability of m⁶A data, transfer learning may be used to enhance TDm6A model performance. We show that TDm6A can learn common and cell-type-specific motifs, some of which are associated with RNA-binding proteins previously reported to be m⁶A readers or anti-readers. In addition, we have used TDm6A to predict m⁶A sites on human long non-coding RNAs (lncRNAs) for selection of candidates with high levels of m⁶A modifications. The results provide new insights into m⁶A modifications on human protein-coding and non-coding transcripts.