TC-6mA-Pred: Prediction of DNA N6-Methyladenine Sites Using CNN with Transformer

Zeeshan Abbas, Mobeen Ur Rehman, Kil To Chong

Abstract: N6-methyladenine (6mA), is amongst the most prevalent post-transcriptional epigenetic changes and is crucial for several cellular functions and disease development. Therefore, a thorough knowledge of cellular processes and other potential functional pathways depends on the proper identification of 6mA modifications. Although we have a number of experimental methods for identifying 6mA-modification sites, computational prediction has emerged as a substitute method as a result of the expensive and labor-intensive nature of experimental methods. Considering this, it is of utmost importance to develop a reliable and effective approach for N6-methy1adenine identification. To categorize genome-wide 6mA locations, numerous computational models have previously been proposed, but there is still potential for advancement in their ability to anticipate 6mA sites. Therefore, we developed a technique using Transformers and neural networks for the accurate prediction of 6mA modification sites in Homo sapiens and Mus musculus genomes and obtained high accuracy of 96.5% and 96.86%, respectively using 5-fold cross-validation technique, and 93.75% on Homo sapiens independent test set. The outcomes demonstrate that the presented model outperforms current approaches in terms of all assessment metrics.