XGBoost-SHAP analysis prioritizes lower-frequency variants predicted to affect HNF4A binding motifs in pancreatic progenitors.

XGBoost-SHAP on seven iPSC-derived pancreatic progenitor single-nucleus ATAC datasets prioritizes a lower-MAF SNP cluster with strong predicted interactions with HNF4A binding motifs.

• Why it matters: Noncoding variants often act through transcription factor binding, but predicting their cell-type-specific regulatory impact is difficult, especially when effects depend on nonlinear feature interactions and variant frequency.
• What they did: The study profiled 93,803 heterozygous SNPs overlapping accessible chromatin, quantified allele-specific effects by binomial tests, trained XGBoost and linear models, and used SHAP to cluster variants by regulatory feature contributions.
• The result: XGBoost outperformed linear models, and the HNF4A-enriched lower-MAF cluster showed highly significant differences in allele-frequency distributions and a stronger relationship between HNF4A-related predictions and CADD-predicted deleteriousness.

AI summary of the abstract · may contain errors · not medical advice · read the paper.