GNNRAI Framework for Multi-Omics Data Integration

This article presents a technical report on the GNNRAI framework, designed for the supervised integration of multi-omics data with biological priors, utilizing explainable graph neural networks (GNNs). The framework aims to address the challenges posed by high-dimensional and heterogeneous multi-omics datasets, which are crucial for identifying biomarkers and predicting disease outcomes. The authors detail the methodology employed in GNNRAI, which includes modeling correlation structures among modality features and leveraging prior knowledge from biological domains. The report demonstrates the application of GNNRAI to Alzheimer’s disease multi-omics data, highlighting its effectiveness in improving prediction accuracy compared to single-omics analyses. The results indicate that the integration of transcriptomics and proteomics data enhances the identification of both known and novel predictive biomarkers. Furthermore, the framework accommodates incomplete data and balances the predictive power of different modalities, showcasing its potential for advancing precision medicine in the context of complex diseases.