Bulk ATAC-seq data of tumour samples result in an averaged signal across different cell-types (cancer, stromal, vascular and immune cells). We propose a deconvolution framework called EPIC-ATAC, which relies on newly identified cell-type specific ATAC-Seq marker peaks and reference profiles for all major cancer-relevant cell-types to predict the proportions of each cell-type. To evaluate EPIC-ATAC, we generated a bulk ATAC-Seq dataset from peripheral blood mononuclear cells (PBMCs) samples, from which the number of cells in each cell-type has been estimated using flow cytometry, as ground truth for cell proportions. The data provided in this Zenodo deposit correspond to: - The raw counts matrix for each peak called in this ATAC-Seq dataset: PBMC_counts.txt - The normalized (TPM-like) counts matrix for each peak called in this ATAC-Seq dataset: PBMC_counts_norm.txt - The cell fractions of each cell type in each sample: PBMC_cell_fractions.txt - The peaks called in each sample using MACS2 (*narrow.peaks): *_normalized.narrowPeak - Bed files listing ATAC-Seq fragments for each sample: *.bed We also evaluated EPIC-ATAC on multiple pseudobulks generated from single-cell ATAC-Seq data. We provide rds files containing the pseudobulks data used in our work for the evaluation of EPIC-ATAC. The rds files are located in the zip file "pseudobulks.zip". The file "additional_data.tar.gz" contains additional files used to generate the reference profiles in EPIC-ATAC and required to run some of the code available on the following GitHub repository: GfellerLab/EPIC-ATAC_manuscript.