We present InsulatorLeak, an open-source computational pipeline that prioritizes non-coding GWAS variants by their predicted disruption of CTCF insulator binding. Using Enformer deep learning predictions (Scaled Average Difference, SAD scores), SuSiE fine-mapping credible sets, and multi-omic colocalization (eQTL, pQTL), we systematically evaluated variants across seven autoimmune diseases — multiple sclerosis, rheumatoid arthritis, inflammatory bowel disease, systemic lupus erythematosus, type 1 diabetes, atopic dermatitis, and psoriasis — spanning more than 2.5 million GWAS participants across 19 datasets. Null-calibrated empirical p-values were computed against locus-matched non-associated variants to control for genomic background. This mechanism-first approach identifies candidate insulator-disrupting variants at established autoimmune loci including IL2RA, IL7R, TYK2, TNFRSF1A, CD58, CD40, BACH2, CLEC16A, and IL12AB. A provisional patent application has been filed with the USPTO (Application No. 63/984,221).
Navya Mihir Shah (Thu,) studied this question.