MultiV_Nm: a prediction method for 2'-O-methylation sites based on multi-view features.

Abstract

As a crucial class of chemical modifications, 2'-O-methylation modification (abbreviated as Nm) is widely distributed in various organisms and plays a very important role in normal cellular physiological activities and the occurrence and development of diseases. Accurate prediction of Nm modification sites can provide important references for the diagnosis and treatment of diseases, as well as identifying for potential drug targets. Aiming at the current problems of unstable performance caused by the use of single features and the need to improve the prediction accuracy of Nm modification sites, this paper proposes MultiV_Nm, a prediction method for Nm sites based on multi-view features. MultiV_Nm extracts the features of Nm sites from multiple dimensions, including sequence features, chemical characteristics, and secondary structure features. By integrating the powerful local feature extraction ability of convolutional neural networks, the ability of graph attention networks to capture global structural information, and the efficient interaction advantage of cross-attention mechanisms for different features, it deeply explores and integrates multi-view features, and finally realizes the prediction of Nm modification sites. The results of cross-validation and independent tests show that this method exhibits significant advantages in key evaluation indicators such as precision, recall, and accuracy, and can effectively improve Nm sites prediction performance. The proposal of MultiV_Nm not only provides a powerful tool for the study of Nm modification but also offers new ideas for predicting other RNA modification sites.