Mapping functional non-coding variation in individual human genomes through haplotyping, multiomics, and deep learning

Most genetic variants in the human genome reside in non-coding regions, where they can perturb regulatory elements, influence gene expression, and contribute to various phenotypes and diseases. However, identifying such functionally relevant genetic variation remains challenging. Here we integrate personal genomics, allele-specific gene regulation, and deep learning predictions to map the impact of non-coding variation in its native allelic and regulatory context. We identify and validate hundreds of cell-type-specific transcription factor binding events disrupted by genetic variants, providing mechanistic insights underlying allele-specific gene regulation. Using this framework, we discover a rare variant that modulates PIK3R5 gene expression by disrupting an OCT2 binding site within a distal enhancer. Our study establishes a generalisable strategy for interpreting non-coding regulatory variation, enabling systematic dissection of variant effects across diverse biological systems and offering an alternative framework to investigate the disease mechanisms.