RNA-protein binding site prediction method and system based on self-attention mechanism

Abstract

The invention discloses an RNA-protein binding site prediction method and system based on a self-attention mechanism. According to the method, a deep learning model is trained through employing the sequence features of an RNA and protein binding site and upstream and downstream, and an RNA-protein interaction binding site is predicted through employing the model; in the coding process of the sequence features, a k-mer embedded coding mode is introduced, the context relationship between adjacent nucleotides is coded, and more effective features are provided for the model; in the feature extraction process, a prediction model is constructed by using a self-attention mechanism, RNA sequence features are focused from the global perspective, and key subsequences are endowed with higher weights so that the network can fully learn key features, and then the model prediction accuracy is improved; and finally, the method provided by the invention is evaluated on a reference data set, and the method is superior to the prior art in the aspect of prediction precision.

Description

technical field [0001] The invention relates to the technical field of biological information, in particular to a method and system for predicting RNA-protein binding sites based on a self-attention mechanism. Background technique [0002] RNA-binding proteins (RBPs) play an important role in gene regulation. Studies have shown that, except for a few RNAs that can function independently in the form of ribozymes, most RNAs participate in gene regulation by combining with proteins to form RNA-protein complexes. RNA binding proteins play key roles in the regulation of life activities such as RNA synthesis, alternative splicing, modification, transport and translation. For example, heterogeneous ribonucleoprotein (HNRNPL) not only directly regulates the alternative splicing of many RNAs, but also regulates the formation of circular RNAs (circRNAs) through back-splicing. [0003] To better understand the function of RNA-binding proteins, researchers need reliable predictive mod...

AI Technical Summary

Problems solved by technology

In addition, in the model stage, the existing models mostly use the combination of convolutional neural network (CNN) and recurrent neural network (RNN), but compared with the self-attention mechanism (self-attention mechanism), the existing model extracts RNA The data features are not sufficient, so the prediction accuracy still needs to be further improved

Images