4mCMS: a deep learning framework based on the multiscale encoding strategy to predict 4mC binding sites

4-Methylcytosine (4mC) represents a crucial DNA base modification, characterized by the addition of a methyl group to the cytosine base. This methylation involves normal gene expression, DNA repair, cell cycle regulation and so on. 4mC binding sites, specific regions within the DNA sequence, engage with 4mC through interactions with proteins or other molecules, influencing gene expression and other cellular functions. Accurate prediction and understanding of the 4mC binding site are of great importance for the in-depth study of the function and regulatory mechanisms of DNA methylation, and for elucidating the mechanisms of cellular processes and disease development. Traditional wet experiments for locating 4mC binding sites, despite their utility, are hindered by high costs and extensive time requirements. Therefore, various computational models have been developed to predict 4mCbinding sites. In this study, we develop a framework named "4mCMS" based on deep learning. The model uses a multi-scale feature encoding strategy that combines different kmers together so that a more comprehensive set of DNA sequence information can be captured. On the C. elegans dataset, the model has an ACC of 0.931, which exceeds the results of the comparative models. The experimental results demonstrate the effectiveness of our proposed model, as we achieve advanced results on the 4mC datasets.