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

Abstract

The identification of cell-type-specific 3D chromatin interactions between regulatory elements can help to decipher gene regulation and to interpret the function of disease-associated non-coding variants. However, current chromosome conformation capture (3C) technologies are unable to resolve interactions at this resolution when only small numbers of cells are available as input. We therefore present ChromaFold, a deep learning model that predicts 3D contact maps and regulatory interactions from single-cell ATAC sequencing (scATAC-seq) data alone. ChromaFold uses pseudobulk chromatin accessibility, co-accessibility profiles across metacells, and predicted CTCF motif tracks as input features and employs a lightweight architecture to enable training on standard GPUs. Once trained on paired scATAC-seq and Hi-C data in human cell lines and tissues, ChromaFold can accurately predict both the 3D contact map and peak-level interactions across diverse human and mouse test cell types. In benchmarking against a recent deep learning method that uses bulk ATAC-seq, DNA sequence, and CTCF ChIP-seq to make cell-type-specific predictions, ChromaFold yields superior prediction performance when including CTCF ChIP-seq data as an input and comparable performance without. Finally, fine-tuning ChromaFold on paired scATAC-seq and Hi-C in a complex tissue enables deconvolution of chromatin interactions across cell subpopulations. ChromaFold thus achieves state-of-the-art prediction of 3D contact maps and regulatory interactions using scATAC-seq alone as input data, enabling accurate inference of cell-type-specific interactions in settings where 3C-based assays are infeasible.

scGrapHiC: deep learning-based graph deconvolution for Hi-C using single cell gene expression

“…We compare scGrapHiC against a comprehensive set of ablation models used as baselines that we have trained and implemented separately to accurately reflect their inputs and architecture. These baselines capture the methodological essence of sequences encoder methods such as Epiphany ( Yang et al 2023 ), Chromafold ( Gao et al 2023 ), and C.Origami ( Tan et al 2022 ) that predict Hi-C contact maps from genomic sequences.…”

“…scRNA-seq+CTCF model: encompasses a set of methods like Chromafold ( Gao et al 2023 ) and C.Origami ( Tan et al 2022 ) that require ATAC-seq with additional structural support through CTCF motif scores to predict cell-specific bulk Hi-C contact maps.…”

“…However, since it did not rely on any cell-type-specific signal, Akita could not predict cell-type-specific interactions. C.Origami ( Tan et al 2022 ), Epiphany ( Yang et al 2023 ), and Chromafold ( Gao et al 2023 ) have extended the sequence encoding framework of Akita to input different cell-type-specific signals (ChIP-Seq or ATAC-Seq) and predict cell-type-specific Hi-C contact maps. These sequence encoder methods make accurate predictions on bulk Hi-C datasets that capture the average cell population signal.…”

scGrapHiC : Deep learning-based graph deconvolution for Hi-C using single cell gene expression

“…However, since it did not rely on any cell-type-specific signal, Akita could not predict cell-type-specific interactions. C.Origami [Ta22], Epiphany [Ya23], and Chromafold [Ga23] have extended the sequence encoding framework of Akita to input different cell-type-specific signals (ChIP-Seq or ATAC-Seq) and predict cell-type-specific Hi-C contact maps. These sequence encoder methods make accurate predictions on bulk Hi-C datasets that capture the average cell population signal.…”

“…scRNA-seq+CTCF model: encompasses a set of methods like Chromafold [Ga23] and C.Origami [Ta22] that require ATAC-seq with additional structural support through CTCF motif scores to predict cell-specific bulk Hi-C contact maps.…”