Zhang ATAC atlas preprocessing for D3 Diffusion Training

Original Data from Paper:Zhang, Kai et al. “A single-cell atlas of chromatin accessibility in the human genome.” Cell 184 (2021): 5985-6001.e19. The code performs 3 steps:1-download-preprocess: Downloads the matrix and converts to a h5ad2-pseudobulk-annotate: Annotates cells by annotated cluster (from paper), and pseudobulks by averaging across cell clusters to get mean expression in each cluster.3-feature-engineering: Performs differential peak analysis to identify cell-type specific peaks, then saves the resulting data as a one-hot encoded .h5 representing cell-type specific accessible sequences for each cell type.The final dataset of this workflow is 3f-dataset.h5