CoRE-ATAC: A deep learning model for the functional classification of regulatory elements from single cell and bulk ATAC-seq data

• DOI: 10.1371/journal.pcbi.1009670
• Publication Date: 2021-12-13T18:35:25Z
• Authors: Asa Thibodeau; Shubham Khetan; Alper Eroglu; Ryan Tewhey; Michael L. Stitzel; Duygu Ucar
• Keywords: Cell type; Chromatin immunoprecipitation; Single-Cell Analysis

Cis -Regulatory elements ( cis -REs) include promoters, enhancers, and insulators that regulate gene expression programs via binding of transcription factors. ATAC-seq technology effectively identifies active -REs in a given cell type (including from single cells) by mapping accessible chromatin at base-pair resolution. However, these maps are not immediately useful for inferring specific functions -REs. For this purpose, we developed deep learning framework (CoRE-ATAC) with novel data encoders integrate DNA sequence (reference or personal genotypes) cut sites read pileups. CoRE-ATAC was trained on 4 types (n = 6 samples/replicates) accurately predicted known -RE 7 40 samples) were used model training (mean average precision 0.80, mean F1 score 0.70). enhancer predictions 19 human islet samples coincided genetically modulated gain/loss activity, which confirmed massively parallel reporter assays (MPRAs). Finally, inferred function aggregate nucleus (snATAC) blood-derived immune cells overlapped functional annotations sorted cells, established the efficacy models to study rare without need sorting. primary reveal individual- cell-specific variation activity. increases resolution maps, critical step studying regulatory disruptions behind diseases.