Test reproducibility of leiden/umap on different systems

This nextflow pipeline runs a jupyter notebook that runs

sc.pp.neighbors
sc.tl.umap
sc.tl.leiden

on a test dataset with a pre-computed latent representation stored in adata.obsm["X_scVI"].

The test-dataset is available in the data directory.

Running the pipeline

There are a few minor things you may need to adjust on your system:

  • The run_pipeline.sh script starts nextflow with a work directory
    and a profile specific to our institute.
  • The clusterOptions is used to submit the jobs to different nodes on our HPC and
    is specific to our queue and node names.
  • The node names are passed on to clusterOptions from a channel defined in main.nf.

Results

The results (tested on 5 nodes, 4 of which have different CPUs) are available from:
https://grst.github.io/scanpy_reproducibility/

GitHub

View Github