Geno-GCN: A Genome-specific Graph Convolutional Network for Diabetes Prediction.

Area of Science:

• Genomics
• Machine Learning
• Bioinformatics

Background:

• Graph convolutional networks (GCNs) show promise but are underutilized in clinical settings for complex diseases like diabetes.
• Predicting Type 2 diabetes risk from whole genome sequencing data presents a significant challenge.

Purpose of the Study:

• To introduce a novel genome-specific graph convolutional network (Geno-GCN) for predicting Type 2 diabetes risk.
• To leverage whole genome sequencing data and incorporate both positive and negative influences of diabetes risk factors.

Main Methods:

• Developed a genome-specific graph convolutional network (Geno-GCN) with a multi-graph aggregator.
• Utilized a negative sample strategy and multi-view aggregators to consolidate risk factor graph influences.
• Assessed Geno-GCN on Australia's largest genome bank.

Main Results:

• Geno-GCN demonstrated superior efficacy and robustness compared to rule-based methods, bioinformatics tools, and other machine learning techniques.
• The method consistently outperformed competitors across all evaluation metrics.
• Geno-GCN showed the closest alignment with actual labels in predicting Type 2 diabetes risk.

Conclusions:

• Geno-GCN is a highly effective and robust method for predicting Type 2 diabetes risk using whole genome sequencing data.
• The proposed approach has significant potential for application in large population studies and clinical settings.
• This work advances the integration of GCNs into genomic medicine for complex disease risk prediction.