snATAC-Express infers Gene Expression from Prioritized Chromatin Accessibility Peaks using Machine Learning

Area of Science:

• Genomics
• Computational Biology
• Immunology

Background:

• Single-cell multiomics offers insights into gene regulation mechanisms.
• Current methods for inferring gene expression from chromatin accessibility have limitations.
• This study proposes a novel approach to model gene regulation considering both positive and negative peak interactions.

Purpose of the Study:

• To develop a machine learning pipeline for integrating single-nuclear multiomic data (transcriptome and chromatin accessibility).
• To model gene expression as a function of ATAC peak intensity, identifying critical regulatory regions.
• To investigate the association of these regulatory regions with genetic variants for immune-related diseases.

Main Methods:

• Developed a machine learning pipeline (snATAC-Express) using random forest regression, XGBoost, Light GBM, and linear regression.
• Applied the pipeline to multiome data from 18 immune cell types across 29 donors (19 with Crohn's disease).
• Utilized coefficient of determination with cross-validation to assess model robustness and predictive accuracy.

Main Results:

• The pipeline identified ATAC peaks significantly contributing to gene expression variation across donors and cell types.
• Models explained 5-40% of transcript abundance variation, using an average of 47% of ATAC peaks.
• Key regulatory peaks were enriched for GWAS variants associated with inflammatory bowel disease and systemic lupus erythematosus.

Conclusions:

• The snATAC-Express pipeline provides a robust method for predicting gene expression from chromatin accessibility data.
• Identified critical ATAC peaks linked to specific autoimmune and inflammatory diseases.
• The developed software, snATAC-Express, is publicly available on GitHub for broader research use.