Deep-learning-based protein function prediction method fusing multiple features

Abstract

The invention provides a deep-learning-based protein function prediction method fusing multiple features. The deep-learning-based protein function prediction method includes the steps: S1, extractingsemantic structure features, subsequence features and network topology structure features of proteins according to sequence information, action information and homologous information of the proteins;and S2, inputting the semantic structure features, the subsequence features and the network topology structure features of the proteins into a pre-trained protein function prediction model, and outputting a classification result, wherein the protein function prediction model comprises: extracting local semantic features of the proteins according to the semantic structure features of the proteins;according to subsequence features of the proteins, extracting denser and higher-level subsequence features of the proteins; fusing the local semantic features of the proteins, denser and higher-levelsubsequence features and network topological structure features of the proteins to obtain protein classification fusion features; and inputting the protein classification fusion features into a function classification module, and outputting a classification result. The deep-learning-based protein function prediction method fusing multiple features obviously improves accuracy of predicting the protein function.

Description

technical field [0001] The present invention relates to the technical field of bioinformatics, in particular to a protein function prediction method based on deep learning fusion of multiple features. Background technique [0002] The function of proteins plays a very important role in the research of biotechnology and medicine, such as the development of new drugs, the development of new crops and the development of synthetic biochemicals such as biofuels. [0003] The early method of predicting protein function was through in vivo and in vitro experiments, including gene knockout, targeted mutation and inhibition of gene expression, etc. These experimental methods required a lot of manpower and time. [0004] In order to alleviate the above problems, some computational methods have been used for protein function prediction, and the development of high-throughput sequencing technology has provided a large amount of effective data for computational methods. Examples include...

AI Technical Summary

Problems solved by technology

Although there are already a large number of computational methods for protein function prediction, the accuracy is generally low

[0005] Protein function prediction is considered to be a huge challenge, mainly for the following reasons: First, a large number of proteins have not yet been annotated. For example, in the UniProt database, less than 1% of proteins have been annotated with functions. method for protein function prediction to further improve the accuracy of protein function prediction

Second, researchers usually use Gene Ontology (Gene Ontology) to label protein functions. Gene Ontology contains more than 40,000 functional categories. Not only does a protein contain multiple functions, but complex biological processes and functions require multiple Proteins act together; that is, protein function prediction is a large-scale, multi-label, multi-classification problem

Third, different data are heterogeneous and complex, so how to use multiple data for protein function prediction is a difficult problem

Images