Iterative clustering

Setting up the environment

Make sure you have the necessary environments installed. For the iterative clustering workflow, you will need the following:

• envs/scanpy.yaml

• envs/rapids_singlecell.yaml (optional, if you have NVIDIA GPU support)

Configuring the workflow

Set up your configuration file

output_dir: data/out
images: images
user_gpu: false

DATASETS:

    my_dataset: # custom task/workflow name

        # input specification: map of module name to map of input file name to input file path
        input:
        clustering:
            file_1: data/pbmc68k.h5ad
            file_2: data/pbmc68k.h5ad # dummy example here for more than 1 file

        # module configuration
        clustering:
          recompute_neighbors: true
          neighbors:
            n_neighbors: 30
            use_rep: X_pca
          recompute_clusters: true
          algorithm: leiden
          resolutions:
            - 1.0
          hierarchy:
            1: 0.1
            3: 0.2
        umap_colors:
          - bulk_labels
          - batch
          - n_genes

Calling the pipeline

Make sure you have set up a runner script to call the Snakemake workflow. The following command will do a dry run of all the steps in the clustering workflow:

bash run_clustering.sh clustering_all -nq

There are also optional evaluation plots that you can call:

bash run_clustering.sh clustering_plot_evaluation_all -nq