# Parse PBMC cytokine: 18 cell types x 12 donors x ~100 cytokines
# Each h5ad file is one cell type with all donors and cytokines.
# Pseudobulk is computed per (donor, cytokine) within each cell type,
# with PBS as the control condition.

dataset: parse

h5ad_dir: /nvme-shared/Data/parse_pbmc/by_cell_type
h5ad_pattern: "*.h5ad"

treatment_col: cytokine
control: PBS

# Within each h5ad (one cell type), stratify by donor.
group_col: donor

# The cell type identity comes from the filename — carry it in output.
file_id_col: cell_type

pdex:
  output_dir: /nvme-shared/Data/parse_pbmc/by_cell_type/pdex_results
  threads: 24
  n_workers: 8  # parallel workers over group_col values

celleval:
  output_dir: /nvme-shared/Data/parse_pbmc/by_cell_type/delta_results
  gene_mode: all_genes
  gene_name_col: null  # use var.index
  target_sum: 3320  # dataset-wide median transcript count