Thank you, I think you are right - I've been conceptualizing (at least for a correlated 2 gene cluster) PC1 capturing the shared main axis of variation, and PC2 capturing the spread away from that main axis (so for positively correlated variables I think you're right that would be the difference)

Ribeiro et. al. has a nice analysis showing eSNPs shared between multiple eGenes are more indeed pleiotropic. We also see this in our data www.nature.com/articles/s41...

🧵5/5 Make allelic proxitropy work for you! Even this simple approach substantially increased colocalizations. We’re excited to see future refinements as we move beyond the “one variant, one gene” paradigm to get a more complete picture of the complexities of local gene regulation 🔬🧬✨

🧵4/5 But are new pcQTLs relevant? We examined associations with GWAS for complex traits and diseases and found over 500 new colocalizations – a 34% increase over single-gene-based analyses! In fact, variants with distributed effects are more likely to be GWAS hits than those affecting just one gene

🧵3/5 In 13 human tissues (using GTEx data), we identified > 12,000 clusters nearby, co-expressed genes, then mapped pcQTLs, finding ~1,400 pcQTLs/tissue. 27% of pcQTL were missed by single-gene QTL approaches! pcQTLs often are smaller distributed effects that we can detect when combined 🤝

🧵2/5 We use principal components (PCs) to summarize nearby genes and map pcQTLs. How do pcQTLs work? 🔍 For correlated genes (A, B, C), standard methods test genotype associations independently along each axis. pcQTLs test for association on a PC axis instead to capture shared regulatory effects 🔗

Excited to share my first PhD paper in the @sbmontgom.bsky.social lab with @tamigj.bsky.social (www.biorxiv.org/content/10.1...)! Standard QTL methods treat each gene independently. But what if a single variant regulates multiple nearby genes at once - what we call “allelic proxitropy”? 🧵 ⬇️