Machine learning prioritises non-coding HNF4A variants active in pancreatic progenitor cells

A PLOS Genetics study applies XGBoost and SHAP analysis to single-nucleus chromatin data, ranking lower-frequency non-coding variants most likely to disrupt HNF4A transcription factor binding in pancreatic development.

Published · AI-drafted summary based on 1 public source
Illustration for generic story
Illustrative image — not from the source article.
Share

Researchers led by Kelly A. Frazer and colleagues at the University of California San Diego have published a study in PLOS Genetics describing a machine-learning framework for identifying non-coding genetic variants predicted to affect binding of the transcription factor HNF4A in pancreatic progenitor cells.

The work uses XGBoost — a gradient-boosted decision tree algorithm — combined with SHAP (SHapley Additive exPlanations) values to disentangle the nonlinear relationships among chromatin accessibility, variant allele frequency, and transcription factor binding affinity. Single-nucleus ATAC-seq data from pancreatic progenitors provided the cell-type-specific chromatin landscape. The analysis prioritised lower-frequency variants predicted to alter HNF4A binding motifs, a finding the authors suggest reflects stronger purifying selection acting on variants with large regulatory effects.

HNF4A encodes a nuclear receptor transcription factor with well-established roles in pancreatic beta-cell function and in maturity-onset diabetes of the young (MODY1). Understanding which non-coding variants plausibly disrupt its activity in a developmentally relevant cell type is a step towards interpreting the large fraction of GWAS signals that fall outside protein-coding sequence.

The study illustrates a broader methodological trend in regulatory genomics: moving from linear models of variant impact towards ensemble approaches that can capture context-dependent, nonlinear effects. The SHAP framework also adds interpretability, allowing researchers to understand which features — chromatin state, motif score, variant frequency — drive each prediction.

Plain-language version

For patients, families, and general readers. Educational only — not medical advice.

Most of our DNA does not encode proteins directly — it contains switches that control when and where genes are turned on. Identifying which DNA changes in these switch regions actually matter for health is technically very difficult. Researchers at UC San Diego have used a type of machine learning called XGBoost, combined with an explanation tool called SHAP, to predict which DNA variants are most likely to disrupt the activity of a protein called HNF4A in developing pancreas cells. HNF4A is important for the cells that make insulin, and certain changes in this gene are linked to a rare inherited form of diabetes. The new analytical framework could help scientists sort important non-coding variants from background noise as they study the genetic basis of conditions like diabetes. This is an educational summary, not medical advice. If anything here raises questions for you, please speak with your GP or a clinical professional.

Sources

Read the original reporting — these are the public sources this summary draws from.

  1. Primary source PLOS Genetics · 2026-11-04
    XGBoost-SHAP analysis prioritizes lower-frequency variants predicted to affect HNF4A binding motifs in pancreatic progenitors

Tags

hnf4a non-coding-variants transcription-factor-binding xgboost shap pancreatic-development regulatory-genomics plos-genetics
Share

About Genetic Current

Educational summaries of public genetics news

Genetic Current is the news section of Evagene, an academic, research, and educational pedigree-modelling platform. Stories are AI-drafted summaries of items from trusted public sources, written for researchers, clinicians, educators, students, genealogists, and patients with an interest in genetics. Summaries are for educational and research purposes only and are not medical advice.

Join the Evagene Alpha Waiting List