Overall, this work has spurred many new hypotheses by pinpointing which genes and contexts are relevant for each disease. As usual, this project was not without it's challenges - I have to give a huge shoutout to my PhD mentors Shamil Sunyaev + Peter Kharchenko and all of our fantastic collaborators

Our method found evidence explaining ~2x the number of GWAS loci compared to pseudobulk eQTLs. We dove deep into several disease-specific findings, especially relating to Parkinson's disease (TRPV2 may be a protective gene, supported by recent studies).

Therefore, our method leverages eQTL models with continuous cell state interactions. While such models are now more commonly used, we specially adapted their output to enable eQTL-GWAS colocalization testing across the cell state space.

Uniquely, scJLIM does NOT require clustering cells before eQTL mapping. Arbitrary clustering cutoffs can kill statistical power for detecting eQTLs when you cluster too finely or too coarsely, and we typically don't know the expected resolution to look at a priori.

This one dramatically increased my awareness of how much we might be wrong with our current view of gene regulation by chromatin - and genesdev.cshlp.org/content/36/1...