Graph neural networks for biomedical insights
summary
Claims a pipeline that fuses transcriptomics and proteomics with a GNN + Transformer architecture on ROSMAP data (~200 paired samples, 20,000 genes 9,000 proteins reduced by GO biological domains). Model outperforms benchmarks on 13 of 16 datasets, uses integrated gradients for attribution, recovers nine known therapeutic targets and proposes 11 novel ones (IPF3 ranked 14), and surfaces a lipid‑metabolism ↔ apoptosis link; notes high p low n, missing modalities, and ongoing work to add imaging and cross‑species translation.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:07Alzheimer's research focus
- 4:21Genetic disease history
- 10:07Public health impact
- 12:00Biomedical data types
- 17:24Data fusion challenges
- 19:53Rosmap study dataset
- 22:01Gene ontology features
- 24:54Biological networks in model
- 26:20GNN for multimodal data
- 28:28Model validation
- 32:27Computational challenges
- 34:13Future research
- 35:17Q&A
related talks
33mNetwork Data Integration Enables Precision Medicine
01h 19mAnalyzing Genetic Networks Using Neural Networks
29mGraph Neural Networks + LLM on
44mNetwork Propagation as a Tool for Deciphering Disease Mechanisms
36mAmazing things that happen with Human-AI synergy
16mAre Neuro-Inspired Multi-Modal Vision-Language Models Resilient to Membership...
39mIdentifying Disease Mechanisms Using Gene Networks as Priors
32mMulti-Omic Integration for Understanding Disease