uniform_batch_correction

Standalone Nextflow pipeline for UniFORM-style batch normalization of cellmeasurement GeoJSON outputs.

Input

Provide a CSV/YAML samplesheet with:

• sample: sample ID
• geojson: path to input cellmeasurement GeoJSON (for feature-level mode)
• ome_tiff: path to input OME-TIFF/TIFF image (for pixel-level mode)
• adata: path to input AnnData .h5ad file (for feature-level matrix mode)

For adata mode, normalization is performed per group inside each .h5ad using an obs column (default: image). Each unique group value is treated as one sample for cohort alignment.

Example CSV is provided in assets/samplesheet.csv.

Usage

nextflow run . \
  -profile conda,large \
  --input assets/samplesheet.csv \
  --outdir results

The large profile targets SLURM large-memory nodes (default queue regular) and increases process resources for large images.

Key parameters

• --run_uniform (default: true)
• --uniform_apply_to (default: geojson; options: geojson, ome_tiff, adata)
• --uniform_num_bins (default: 1024)
• --uniform_min_value (default: 1.0)
• --uniform_exclude_pattern (default: ^(kronos_|emb_))
• --uniform_output_suffix (default: _uniform)
• --uniform_pixel_output_suffix (default: _unifrom)
• --uniform_pixel_sample_size (default: 200000)
• --uniform_adata_group_by (default: image)
• --uniform_adata_sample_size (default: 200000)
• --uniform_adata_target (default: all; use cell_mean to target only *_Cell_Mean features)
• --uniform_adata_filter_column (default: empty; e.g. statistic)
• --uniform_adata_filter_regex (default: empty; only matching features are normalized)
• --uniform_generate_plots (default: true)
• --uniform_qc_top_n_keys (default: 12)
• --uniform_qc_max_heatmap_keys (default: 40)

Normalized GeoJSON files are published to results/uniformnormalize/. Normalized AnnData files are published to results/uniformnormalize/. QC files are published to results/uniformnormalize/qc/.