GENRAD: Genomics and Radiomics Heterogeneous Graph Neural Network for Graph-Level Classification in Alzheimer's Disease
Keywords: Heterogeneous Graph Neural Network (GNN), Alzheimer's Disease, Multimodal data fusion, Genomics, Radiomics, Node-level classification, Multi-scale graph representation, Explainable AI
TL;DR: GENRAD, a heterogeneous graph neural network for Alzheimer's Disease classification, integrates genomic and radiomic data, achieving 91.70% accuracy and providing biologically meaningful insights via multi-scale graph representations.
Abstract: Alzheimer’s Disease (AD) poses multifaceted challenges due to its neurodegenerative nature driven by complex genomic, radiomic, and structural interactions. Understanding these complex relationships is pivotal for advancing diagnostic and therapeutic approaches. Current models struggle to effectively integrate multimodal data for AD, limiting their predictive accuracy and biological interpretability. Thus, there is a pressing need for models that can seamlessly fuse genomic and radiomic data to provide a holistic understanding of AD pathology. We introduce GENRAD, a novel heterogeneous graph neural network (GNN) that integrates multimodal genomic and radiomic data for graph-level classification in AD by representing patients, genes, and brain structures as distinct nodes and implementing advanced message-passing techniques. The benefits of GENRAD are fourfold: (1) It enables multimodal fusion of genomic and radiomic data, uncovering biologically meaningful insights missed by single-modality models. (2) Its adaptive multi-scale graph representations model interactions at various biological scales, capturing complex relationships essential for understanding AD pathology. (3) GENRAD incorporates explainable AI techniques, providing detailed analysis of key genomic markers and brain regions associated with AD. (4) GENRAD performs unsupervised clustering of genes, allowing the identification of functionally related biological pathways, thus empowering clinicians with actionable insights for personalized treatment strategies. GENRAD demonstrates superior classification accuracy in identifying AD-related patterns compared to existing machine and deep learning models, achieving an accuracy of 91.70%.
Primary Area: learning on graphs and other geometries & topologies
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4409
Loading