Geno-GCN: A Genome-specific Graph Convolutional Network for Diabetes Prediction.
Journal:
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
PMID:
40039720
Abstract
Drawing inspiration from convolutional neural networks, graph convolutional networks (GCNs) have been implemented in various applications. Yet, the integration of GCNs into clinical settings, particularly in the context of complex health conditions like diabetes, remains distant. In this paper, we introduce a genome-specific graph convolutional network (Geno-GCN) with a multi-graph aggregator to predict the risk of developing Type 2 diabetes based on whole genome sequencing data. Geno-GCN consolidates both positive and negative influences from graphs formulated from diabetes risk factors. This is achieved through a negative sample strategy combined with multi-view aggregators. We assessed Geno-GCN using Australia's largest genome bank and benchmarked it against rule-based methods, bioinformatics tools, and other state-of-the-art machine-learning techniques. The results demonstrated the superior efficacy and robustness of our method, which consistently outperformed competitors across all evaluation metrics. Geno-GCN also exhibited the closest alignment with actual labels, showcasing its potential in large population studies.