Kidzoi Enables Cell-Type-Specific Regulatory Activity and Variant Effect Prediction in Kidney

Kidzoi, a kidney cell-type-specific S2F model, improves regulatory variant effect prediction by 10-23% over existing models.

• Why it matters: Current S2F models are limited by their training datasets and cannot fully capture the diversity of cell types and conditions in the kidney, hindering precise gene regulation understanding and variant prioritization.
• What they did: The authors developed Kidzoi through transfer learning from Borzoi, using single-nucleus chromatin accessibility data from human kidney tissue, and systematically tested its performance on predicting chromatin accessibility and GWAS causal variants.
• The result: Kidzoi effectively prioritizes kidney function-associated causal variants with cell-type specificity, and a multi-task version maintains performance while reducing computational costs, enabling more efficient variant analysis.