SICM6A: Identifying m6A Site across Species by Transposed GRU Network

N6-methyladenosine (m6A) is the most prevalent cross-species RNA methylation modification and plays a pivotal role in various biological processes. The biochemical methods to find m6A sites are expensive and time-consuming, and the false positive rate of identified sites is high relatively. Meanwhile, the current computations are complex, and the prediction performance is relatively low both on little data sets and large data sets. This paper, at this point, presents a deep learning model with a transposed operation in the middle of GRU layers, SICM6A, for identifying m6A sites across-species. It adopts the mixed precision training manner to improve the speed and performance, and predicts m6A sites only by directly reading the 3-mer encoding of the m6A short sequence. The cross-validation and independent test verification show SICM6A is more accurate than the state-of-the-art methods. This, therefore, makes SICM6A provide new idea for predicting other modification sites of RNA sequences. The prediction software SICM6A is on github (https://github.com/lwzyb/SICM6A).