Deep Learning-based Identification of DNA-N4 Methylcytosine Modification Sites

Abstract

DNA modification is closely related to the expression genetics of many organisms, therefore, the prediction of DNA modification sites is particularly important. In this paper, we use deep learning techniques to identify and predict DNA N4-methylcytosine modification sites, and the main work is as follows. Feature encoding using k-spacer nucleic acids to encode a 41 bp long DNA sequence as a (41×9) dimensional vector. Recognition prediction based on multi-headed attention mechanism and GRU neural network. Firstly, the encoded data are extracted and downscaled; secondly, the importance distribution of 4mc loci and each nucleotide in the sequence are further extracted adaptively using the multi-headed attention mechanism; then the GRU network is used to capture the long dependencies in the whole importance distribution; finally, a new prediction model of 4mc loci is constructed using two fully connected layers, and its recognition accuracy is significantly improved compared with other basic machine learning models. The recognition accuracy is improved compared with other basic machine learning models.