Construction of an Exosome Antioxidant Pathway Recognition Algorithm by Integrating Bioinformatics Mining with Graph Neural Networks

Abstract

It is very important to identify and understand the antioxidant pathways in exosomes efficiently, so as to deeply explore the biological functions of exosomes and develop related therapeutic methods. This article aims to construct an exosome antioxidant pathway identification algorithm that combines bioinformatics mining and graph neural network (GNN). In this article, the characteristics of gene expression, protein interaction and metabolic pathway are extracted, and GNN training is optimized by random node sampling and random walk sampling. Then, an efficient and accurate prediction model is proposed by combining the improved graph convolution network (GCN) and multi-task convolutional neural network (CNN). The results show that the accuracy of the proposed algorithm on the test set is 89.7%, and the F1 score is 88.7%, which is obviously superior to the traditional method. In addition, the model successfully identified several key molecules, such as SOD1 and GPX1, and their functional annotations were highly consistent with the existing literature. These findings can verify the effectiveness of the algorithm and provide a new perspective for understanding the biological function of exosomes in antioxidant mechanism.