scEpiLock: A Weakly Supervised Learning Framework for cis-Regulatory Element Localization and Variant Impact...

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Single-cell transposase-accessible chromatin sequencing (scATAC-seq) offers high-resolution cellular heterogeneity dissection.
• Ultra-high missingness in scATAC-seq leads to broad peak definitions, limiting functional region identification and variant impact interpretation.
• Accurate identification of functional genomic regions and cell-type-specific variant impacts is crucial for understanding gene regulation and disease mechanisms.

Purpose of the Study:

• To develop a weakly supervised learning method, scEpiLock, for precise identification of core functional regions from scATAC-seq data.
• To quantify variant impacts in a cell-type-specific manner.
• To improve peak boundary definitions and enhance the interpretation of variant effects in regulatory elements.

Main Methods:

• scEpiLock employs a multi-label classifier with a deep convolutional neural network to predict chromatin accessibility.
• A weakly supervised object detection module, utilizing gradient-weighted class activation mapping (Grad-CAM), refines peak boundary definitions.
• The method quantifies cell-type-specific variant impacts within defined peak regions.

Main Results:

• scEpiLock achieved an AUC of ~0.9 and AUPR above 0.7 on various scATAC-seq datasets.
• The object detection module condensed coarse peaks to one-third of their original size while maintaining higher conservation scores.
• Application to brain scATAC-seq data identified GWAS variants impacting regulatory elements near Alzheimer's disease risk genes.

Conclusions:

• scEpiLock effectively refines peak boundary definitions and enhances the identification of functional genomic regions in scATAC-seq data.
• The method provides accurate cell-type-specific quantification of variant impacts, crucial for interpreting genetic associations.
• scEpiLock demonstrates significant potential for generating cell-type-specific biological insights in complex disease studies, such as Alzheimer's disease.