Synthetic lethal interaction prediction method based on heterogeneous graph convolutional neural network

Abstract

The invention relates to the field of data mining in bioinformatics, in particular to a synthetic lethal interaction prediction method based on a heterogeneous graph convolutional neural network. The method mainly comprises the following steps: (1) collecting known synthetic lethal correlation data, gene GO information data and gene PPI data; (2) carrying out gene GO similarity analysis, measuring GO function similarity between genes by utilizing a semantic gene function similarity measurement algorithm, and constructing GO function similarity-based features of the genes; (3) constructing genes based on PPI characteristics, constructing a correlation network between proteins by utilizing protein correlation data, and obtaining the characteristics of each gene based on the protein correlation network in a random walk mode; (4) constructing an adjacent matrix by using synthetic lethal correlation data, and fusing domain features of the gene based on GO function similarity features and PPI features; and (5) constructing a synthetic lethal pair prediction model based on the graph convolutional neural network, predicting potential synthetic lethal interaction, and obtaining a final result.

Description

technical field [0001] The invention relates to the field of data mining in bioinformatics, in particular to a synthetic lethal interaction prediction method based on a heterogeneous graph convolutional neural network. Background technique [0002] Synthetic lethality (SL) plays a crucial role in tumor therapy because it recognizes specific targeted genes to kill tumor cells without interfering with normal cells. Since the verification of SL pairs by a high-throughput wet experimental setup is often expensive and time-consuming, it also faces various challenges. Therefore, in recent years, many researchers have tried to significantly reduce the cost and time of identifying SL pair interactions by using computational methods to verify and predict SL pairs. [0003] Early computational methods for SL pair recognition can be divided into two categories: methods based on big data and data mining. This data-driven approach includes biological network topology methods, data mini...

AI Technical Summary

Problems solved by technology

However, feature-based machine learning methods often map genes in isolation into isolated latent representations, ignoring neighbor gene information

At the same time, with the improvement of major public data sources, more and more biological data are available, and it has become a big challenge to select which heterogeneous data to process to obtain more effective feature representation.

Images