软件导刊2025，Vol.24Issue(5)：28-34,7.DOI:10.11907/rjdk.241221
基于图神经网络的疾病表型预测方法研究
Research on Disease Phenotype Predicting Method Based on Graph Neural Network
摘要
Abstract
The human gut microbiota is closely related to human health,and the occurrence of many diseases is accompanied by dysregulation in the gut microbiota.Accurately predicting host phenotypes from gut microbiome data is of great significance for understanding the occurrence and development of diseases.Machine learning methods have shown strong capabilities in predicting phenotypes from microbiome data.Howev-er,such as sparsity,high dimensionality,and correlations among features in microbiome data limit the accuracy and reliability of disease phe-notype prediction based on machine learning.Therefore,the paper proposes a graph classification framework named PAMNC based on microbi-ome correlation networks to learn to classify microbiome correlation networks.Firstly,it learns the phylogenetic relationships between sample compositions to enhance the information of original abundances;secondly,it constructs a set of microbiome correlation networks using Bayesian compositional-aware correlation inference;finally,it applies graph neural network-based methods for disease phenotype prediction.Experi-mental results on six public metagenomic datasets and nine advanced phenotype prediction methods show that PAMNC consistently outperforms other methods.Particularly,on the Cirrhosis and IBD disease datasets,the PAMNC method achieves AUC values of 0.940 2 and 0.911 2,re-spectively.关键词
机器学习/微生物组分析/微生物相关网络/图表示学习/表型预测Key words
machine learning/microbiome analysis/microbial relation network/graph representation learning/phenotype prediction分类
信息技术与安全科学引用本文复制引用
石凯,柳乔晖,冀庆荣,赵鹏阳..基于图神经网络的疾病表型预测方法研究[J].软件导刊,2025,24(5):28-34,7.基金项目
国家自然科学基金项目(62162019) （62162019）
