Searching For Causal Genetic Mechanisms Across Human Populations

on YouTube ↗33m

" summary

Proposes a concrete pipeline that builds knockoff framework variables per population, runs Lasso-style importance scores, and tests conditional independence given the rest of the genome across environments. Uses a simulation with five real populations and UK Biobank subgroups, shows multi-environment knockoffs control causal FDR where single-population or naive pooling fails, and reports more discoveries with higher external validation rates when requiring consistency across environments.

AI-written summary · about

" chapters (15)

1. 0:00Intro
2. 2:09One vantage point: gene mapping for complex traits
3. 3:15GenomeWide Association Studies and causality
4. 4:35GenomeWide Association Studies Standard analysis (simplified)
5. 5:31GWAS and human populations
6. 6:35An alternative approach
7. 12:35Covariate shifts across human populations
8. 14:53Testing consistent conditional hypotheses with FDR control
9. 16:47Gaussian dummies
10. 17:47Permuted dummies
11. 19:21Constructing knockoff scores environment by environment
12. 20:14Testing the invariant hypotheses
13. 21:17Genetic simulation
14. 29:01UK biobank data analysis - external validation of results
15. 30:49Today's talk, GWAS, human populations & causality