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

@article{frazer2026xgboostshap,
  title = {{XGBoost-SHAP analysis prioritizes lower-frequency variants predicted to affect HNF4A binding motifs in pancreatic progenitors}},
  author = {Kelly A. Frazer and Benjamin Henson and Jennifer P. Nguyen and Wilfredo G. Gonzalez Rivera and Matteo D’Antonio},
  journal = {PLOS Genetics},
  year = {2026},
  month = nov,
  doi = {10.1371/journal.pgen.1012361},
  url = {https://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1012361},
  note = {CC BY 4.0},
}

TY  - JOUR
TI  - XGBoost-SHAP analysis prioritizes lower-frequency variants predicted to affect HNF4A binding motifs in pancreatic progenitors
AU  - Kelly A. Frazer
AU  - Benjamin Henson
AU  - Jennifer P. Nguyen
AU  - Wilfredo G. Gonzalez Rivera
AU  - Matteo D’Antonio
PY  - 2026
DA  - 2026/11/04
JO  - PLOS Genetics
DO  - 10.1371/journal.pgen.1012361
UR  - https://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1012361
AB  - Background: Noncoding genetic variants underlie many complex traits, yet predicting their regulatory impact in cell type-specific contexts remains a major challenge. Transcription factor binding is a primary mechanism through which noncoding variants influence gene regulation, but the relationships among chromatin accessibility, variant frequency, and transcription factor binding are complex and nonlinear, limiting the performance of traditional linear models. Methods: We generated single-nucleus ATAC sequencing data from seven induced pluripotent stem cell-derived pancreatic progenitor cell samples and identified 93,803 heterozygous single nucleotide polymorphisms overlapping accessible chromatin regions. Allele-specific effects (ASE) were quantified for each variant via binomial tests. We trained XGBoost, logistic regression, and regularized logistic regression models to predict allele-specific effects via chromatin read depth, peak score, variant minor allele frequency (MAF), and transcription factor binding prediction for five pancreatic transcription factors: HNF4A, HNF1A, NEUROD1, NKX6-1, and PDX1. Model performance was assessed by the area under the receiver operating characteristic curve (AUC) via fivefold cross-validation as well as using test and validation SNP sets. Shapley (SHAP) value analysis was used to interpret model predictions and to cluster variants on the basis of their regulatory feature contributions. Results: XGBoost significantly outperformed linear models, particularly at stringent allele-specific effect thresholds, consistent with its ability to capture nonlinear interactions among features. Chromatin read depth, peak score, and minor allele frequency were the strongest predictors. Stratification by read-depth and MAF deciles showed that XGBoost retained better overall predictive performance across deciles than the linear models, indicating that its advantage was not explained by these variables alone. Dimensionality reduction of variants on the basis of Shapley value profiles revealed seven clusters with distinct chromatin states and variant allele frequencies. These findings were supported by highly significant pairwise differences in allele frequency distributions across the seven clusters. One cluster was enriched for lower-MAF SNPs showing strong predicted interactions with HNF4A binding motifs. Because MAF and its SHAP contribution were inputs to the embedding, the lower-MAF identity of this cluster was expected; its high ASE enrichment and distinctive HNF4A interaction structure were the biologically informative findings. Combined Annotation Dependent Depletion (CADD) analyses further supported a stronger relationship between HNF4A-related predictions and predicted deleteriousness in this cluster. Conclusions: Integration of single-nucleus chromatin accessibility data with an interpretable XGBoost-SHAP framework provides a method for prioritizing noncoding variants and identifying nonlinear regulatory relationships in a specific developmental context. Within this framework, the stronger relationship between HNF4A-related predictions and CADD-predicted deleteriousness in the lower-MAF cluster is consistent with selective constraint on HNF4A-bound regulatory elements and suggests a hypothesis for how lower-frequency noncoding variation may contribute to intrafamilial phenotypic variability in HNF4A-related Maturity-Onset Diabetes of the Young (MODY1). Author summary: Most genetic variants linked to common diseases are located outside of coding genes, in regulatory regions that control the cell types and timing of gene expression. Identifying the noncoding variants that actually change gene regulation, and determining the underlying functional mechanism, remains difficult. Here we studied pancreatic progenitor cells, the developmental precursors of the insulin-producing cells of the pancreas, using single-nucleus ATAC-seq to measure chromatin accessibility in seven individuals. We built an interpretable machine-learning model (XGBoost) to predict which variants show allele-specific effects on chromatin accessibility, and used Shapley (SHAP) values to understand what the model learned. XGBoost outperformed linear models, consistent with its ability to capture nonlinear interactions among features. The model relied most heavily on chromatin accessibility measures, which reflect both biological accessibility and the read depth available to detect allele-specific effects. Grouping variants by their SHAP profiles revealed distinct predictive categories. We identified one particularly interesting group composed of variants with relatively lower allele frequencies whose predicted effects were associated with HNF4A sequence binding motifs. Although this group’s lower allele frequency was not an unexpected finding because allele frequency was one of the inputs used to cluster variants, its distinctive HNF4A interaction structure and high enrichment for allele-specific effects were biologically informative. CADD analyses further supported a stronger relationship between HNF4A-related predictions and predicted deleteriousness in this group. HNF4A is a transcription factor essential for pancreatic development and mutated in a form of monogenic diabetes (Maturity-Onset Diabetes of the Young, MODY1). These findings may provide insights into phenotypic variability among family members segregating MODY1. Our results show that even small single-cell datasets, combined with interpretable machine learning, can prioritize candidate regulatory variants for experimental follow-up.
SN  - 1553-7404
ER  -