Thread by @kanishkadey

Excited to share our latest preprint, as part of ENCODE4, on a consensus variant-to-function (cV2F) score for functionally prioritizing variants for complex disease - led by Tabassum Fabiha and co-mentored with Alkes Price.
biorxiv.org/content/10.110…

The combination of ENCODE4 variant-level experimentally and computationally predicted function + element-level annotations, conservation, LD show the highest predictive accuracy for GWAS fine-mapped variants.

cV2F shows better heritability signal for complex traits and better concordance with MPRA implicated variants compared to other variant-function and pathogenicity scores.

cV2F model trained on GWAS data has better transferability to other functional data - reporter assays, QTLs, CRISPR (likely function of limited data size, quality, and tissue coverage).

cV2F-informed fine-mapping results in 14.3% more confidently fine-mapped (PIP > 0.95) variants than non-functionally informed fine-mapping (versus 10.7% for PolyFun+SuSIE baseline-LD model)

Tissue/cell-line-specific cV2F scores show unique heritabilty signal in matched traits conditional on tissue-agnostic cV2F, as well specifically higher enrichment in reporter assays in matched cell lines.

We thank all ENCODE4 consortium collaborators - Ivy Evergreen, @soumyakundu_ , @panushri25 , Sergey Abramov, Sasha Boytsov, @KariStrouse , @ksmualim , @argschwind , @GaspardKerner , John Butts, Katherine Dura, Weixiang Fang, @ThahminaAli

and ENCODE4 PIs - @MooreJillE @ZhipingWeng @julirsch @jihk99 @jeffvierstra Tim Reddy, @sbmontgom , @jengreitz , @anshulkundaje @r_tewhey