KGWAS is a novel geometric deep learning method that leverages a massive functional knowledge graph across variants and genes to significantly improve detection power in small-cohort GWASs. Thank you, Kexin Huang and Martin Jinye Zhang, for joining me on this one.Preprint:https://www.medrxiv.org/content/10.11...GitHub Page:https://github.com/snap-stanford/KGWASOfficial Website:https://kgwas.stanford.edu/

📖 Video Chapters:0:00 - Introduction to Rare Disease Genetics Problem0:54 - Welcome & Paper Overview1:22 - What is KGWAS? The Elevator Pitch2:41 - Why Finding More GWAS Hits Matters4:46 - Origin Story: Collaboration with GSK6:09 - Knowledge Graph Architecture Overview7:36 - Building the Variant-Gene-Pathway Network9:55 - Handling Linkage Disequilibrium (LD) Challenges13:55 - Training the Large-Scale Graph Neural Network14:16 - Model Training Requirements and Scalability15:22 - Open Source Availability and Accessibility16:05 - Scaling to Whole Genome Analysis17:12 - Figure 2: Validation Through Simulations18:24 - UK Biobank Downsampling Experiments20:04 - Precision-Recall Performance Results21:00 - Figure 3: Real-World Disease Applications21:38 - Ulcerative Colitis Case Study Example22:48 - Experimental Validation of Ulcerative Colitis Discovery24:03 - Myasthenia Gravis Case Study Analysis26:10 - Knowledge Graph Component Analysis and Ablation Studies27:54 - Tissue-Specific vs Context-Agnostic Approaches32:03 - Figure 4: Network Interpretation and Attention Mechanisms36:30 - Alzheimer's Disease Network Visualization40:01 - Drug Development Applications and Implications41:15 - Post-GWAS Era Applications and End-to-End Solutions43:01 - Figure 5: Compatibility with Existing GWAS Tools48:16 - Population Diversity and Cross-Ancestry Applications49:41 - Future Directions and Technical Improvements52:16 - Input/Output Requirements and Computational Resources54:28 - Website Demo and Interactive Features56:59 - PhD Student Advice and Research Philosophy59:36 - Closing Remarks