Cell Type Prediction

Cell Type Prediction#

Cell Type Prediction#

This module enables automated cell type annotation of single-cell RNA-seq datasets using pre-trained models. This allows for:

  • Consistent cell type annotation across studies

  • Probabilistic cell type assignments with confidence scores

  • Majority voting and over-clustering analysis for robust predictions

The module currently supports:

  • CellTypist: Automated cell type annotation using pre-trained models

Environments#

The following environments are needed for cell type prediction:

Configuration#

DATASETS:
  test:
    input:
      celltype_prediction:
        preprocessed: test/input/preprocessing/dataset~all/file_id~pbmc/preprocessed.zarr
    celltype_prediction:
      reference_label: bulk_labels
      counts: layers/counts
      is_normalized: false
      celltypist:
        params:
          majority_voting: true
          over_clustering: bulk_labels
        models:
          - Healthy_COVID19_PBMC
          - Immune_All_Low

Input#

The input AnnData object should contain the single-cell RNA-seq data to be annotated. Reference cell type labels can optionally be provided for evaluation and visualization purposes.

Configuration Parameters#

  • counts (default: 'X'): Which data layer to use from the AnnData object

  • is_normalized (default: true): Boolean flag indicating whether the input data is already normalized

    • CellTypist expects log-normalized data, so this parameter controls preprocessing

  • reference_label (optional): Column name in .obs containing reference cell type labels for comparison and visualization

CellTypist Parameters (celltypist)#

Configuration for CellTypist cell type prediction:

  • models: List of pre-trained model names to use for prediction (required)

    • Available models include Healthy_COVID19_PBMC, COVID19_HumanChallenge_Blood, Immune_All_Low, etc.

    • Multiple models can be applied to the same dataset

  • params: Model parameters (optional)

    • majority_voting: Enable majority voting across over-clustering results (default: false)

    • over_clustering: Column name in .obs for over-clustering analysis (optional)

Note: CellTypist models are trained on specific tissue types and cell populations. Choose models appropriate for your data type (e.g., PBMC, immune cells, etc.).

Output#

CellTypist#

The cell type prediction workflow produces the following outputs:

  • <out_dir>/celltype_prediction/dataset~<dataset>/file_id~<file_id>.zarr: Annotated AnnData object containing:

    • Direct predictions (obs['celltypist_<model>:predicted_labels']): Primary cell type predictions

    • Majority voting results (obs['celltypist_<model>:majority_voting']): Consensus predictions (if enabled)

    • Over-clustering results (obs['celltypist_<model>:over_clustering']): Fine-grained clustering results (if specified)

    • Confidence scores (obs['celltypist_<model>:conf_score']): Prediction confidence values

  • <out_dir>/images/: Visualization plots comparing predictions with reference labels (if provided)