Doublet Detection

Doublet detection

This module runs per-batch doublet detection using one or more callers (currently doublet_detection and scrublet). The pipeline runs doublet callers within each library/batch and writes scores and predictions back into the AnnData object.

Environments

The following environments are useful for running the module. Install only the ones you need.

• scanpy

• (caller-specific envs may be required for scrublet or doublet_detection)

Configuration

DATASETS:
  Lee2020:
    input:
      doublets:
        Lee2020: test/input/load_data/harmonize_metadata/Lee2020.zarr
    doublets:
      counts: X
      batch: donor
      chunk_size: 10_000

  test:
    input:
      doublets:
        test: test/input/pbmc68k.h5ad
        test2: test/input/pbmc68k.h5ad
    doublets:
      counts: layers/counts
      methods:
        - scrublet
        - doubletdetection

defaults:
  datasets:
    - test
    - Lee2020

• counts: Slot in anndata that contains raw (unnormalized) counts. Examples: X, raw/X, or layers/<layer_name>.

• batch: Column in obs that contains batch/library IDs. Doublet detection is executed separately per batch.

• chunk_size: Number of cells used to group batches for more efficient parallel processing. Default: 100_000.

• methods: Optional list of doublet callers to run for this dataset (e.g. ['scrublet', 'doubletdetection']). If omitted, pipeline defaults apply.

Note: counts is resolved relative to the input object (e.g., anndata.X or anndata.layers). Methods are run per-batch; choose batch to reflect library-level grouping so cross-library doublets are not considered.

Output

Results are written into the AnnData .obs and to summary files/plots:

• <out_dir>/doublets/dataset~<datasets>/file_id~<file_id>.zarr — Updated AnnData with caller-specific score and prediction columns in .obs, for example:

  • scrublet:

    • doublet_score

    • predicted_doublet

  • doublet_detection:

    • doublet_score_doubletdetection

    • doubletdetection_prediction

• <image_dir>/doublets/dataset~<datasets>/file_id~<file_id>/doublet_summary.tsv — Summary table with per-batch and per-method statistics.

• <image_dir>/doublets/dataset~<datasets>/file_id~<file_id>/plots/ — Visualizations of score distributions and per-batch results (e.g., histograms, per-batch summary plots).

If additional callers are enabled, corresponding score/prediction columns will be added to .obs using method-specific names to avoid collisions.