ChromaFold predicts the 3D contact map from single-cell chromatin accessibility.

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Identifying cell-type-specific 3D chromatin interactions is crucial for understanding gene regulation and non-coding variant function.
• Current chromosome conformation capture (3C) technologies struggle with low cell input, limiting resolution.
• Interpreting the functional impact of non-coding genetic variants remains a challenge.

Approach:

• Developed ChromaFold, a deep learning model predicting 3D contact maps and regulatory interactions from single-cell ATAC sequencing (scATAC-seq) data alone.
• ChromaFold utilizes pseudobulk chromatin accessibility, metacell co-accessibility, and predicted CTCF motif tracks as input.
• Employs a lightweight architecture for efficient training on standard GPUs.

Key Points:

• ChromaFold accurately predicts 3D contact maps and peak-level interactions across diverse cell types after training on paired scATAC-seq and Hi-C data.
• Outperforms existing methods using bulk ATAC-seq when CTCF ChIP-seq data is included and shows comparable performance without it.
• Fine-tuning enables deconvolution of chromatin interactions within cell subpopulations in complex tissues.

Conclusions:

• ChromaFold achieves state-of-the-art prediction of 3D chromatin interactions using only scATAC-seq data.
• Enables accurate inference of cell-type-specific interactions in scenarios where traditional 3C-based assays are not feasible.
• Provides a powerful tool for dissecting gene regulation and interpreting disease-associated variants at single-cell resolution.