SICD6mA: Identifying 6mA Sites using Deep Memory Network

Background DNA N6-methyladenine (6mA) is a kind of epigenetic modification in prokaryotes and eukaryotes, which involves multiple biological processes, such as gene regulation and tumorigenesis. Identifying 6mA contributes to understand its regulatory role. Therefore, to satisfy the needs of large-scale preliminary screening, it is necessary to develop the high-quality computational models for the rapid identification of 6mA sites. However, the existing calculation approaches are mostly specific to rice, and they have not been extensively applied to human genome. Results This study proposed a classification method of deep learning based on the memory mechanism named SICD6mA. In addition, the large benchmark datasets were constructed for human and rice, respectively, which integrated the recently reported 6mA sites. According to the evaluation results, SICD6mA displayed favorable robustness during cross-validations, which achieved the area under the curve (AUC) values of 0.9824 and 0.9903 for Human and Rice’s genomes in independent test evaluations, separately. Conclusions The successful prediction rate of 6mA sites on cross-species genomes exhibited higher accuracy than that of the state-of-the-art methods. For the convenience of experimental scientists, the user-friendly tool SICD6mA was developed to predict the cross-species 6mA sites, thereby accelerating and facilitating future cross-species genome research.