Briefly, we analyzed blood trait GWAS, eQTLs from blood cell types, and CRE (enhancer) CRISPRi+scRNA-seq data from K562s, primarily from Gasperini et al. and our own Morris et al. (with @johnomix.bsky.social). We were then able to analyze the overlap of discoveries in 882 GWAS CREs. - ThreadSky

tuuliel.bsky.social • 21 days ago

Briefly, we analyzed blood trait GWAS, eQTLs from blood cell types, and CRE (enhancer) CRISPRi+scRNA-seq data from K562s, primarily from Gasperini et al. and our own Morris et al. (with @johnomix.bsky.social). We were then able to analyze the overlap of discoveries in 882 GWAS CREs.

Comments

tuuliel.bsky.social•21 days ago

Target genes do overlap, but not so much. This is partly due to power: sc-eQTLs contribute little, eQTLs struggle with low-freq variants, CRISPRi with low expr genes. CRISPRi (but not eQTL) power is low in non-overlapping loci, indicating that bigger CRISPRi studies -> better convergence.

tuuliel.bsky.social•21 days ago

eGenes and cGenes have different properties too: cGenes are fewer, proximal, more constrained and more buffered, while eQTL mapping (especially across many cell types) finds more genes per locus, including distal ones. This is consistent with previous empirical work and theoretical work on eQTLs.

tuuliel.bsky.social•21 days ago

The 💰 question: what about “true” causal gene discovery? We compiled a list of 421 gold-standard genes, of which 53 were discovered as eGenes and 23 as cGenes - and only 9 with both! A higher proportion of cGenes are gold-standard genes, compared to eGenes or Hi-C targets.

tuuliel.bsky.social•21 days ago

Analysis of the loci where the gold-standard gene was missed looks pretty grim: eQTLs find other i.e. misleading genes, while CRISPRi often lacks power. "Wow they both suck, but in totally different ways" was my first reaction to this plot.

tuuliel.bsky.social•21 days ago

The analysis of gold-standard genes in these GWAS loci shows how much you'd miss using only one approach, and how startlingly orthogonal the discoveries are. The patterns are highly variable between the loci. It depends on your downstream application if you prefer higher sensitivity vs specificity.

tuuliel.bsky.social•21 days ago

Overall, this provides important insights in cis-target gene discovery from eQTLs and CRISPRi. I think we're ultimately going to need both tools, as both have fundamental weaknesses: eQTL's bias against constrained genes, and difficulty of applying CRISPR to diverse cell types.

Comments

Posting Rules

Reply