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
clusterOptionsis 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
clusterOptionsfrom a channel defined inmain.nf.
Results
The results (tested on 5 nodes, 4 of which have different CPUs) are available from: https://grst.github.io/scanpy_reproducibility/
3 Oct 24, 2022
8 Dec 14, 2022
1 Sep 25, 2022
2 Jan 12, 2022
1 Feb 2, 2022
1 Dec 9, 2022
16 Aug 18, 2022
4 Feb 9, 2022
91 Dec 28, 2022
1 Jul 11, 2022
3 Nov 08, 2022
71 Dec 19, 2022
33 Aug 20, 2022
2 Jan 25, 2022
3 Dec 10, 2022
2 Oct 27, 2021
1 Mar 15, 2022
13 Oct 09, 2022
1 Jan 27, 2022
1 Feb 18, 2022
1 Jul 19, 2022
21 Dec 20, 2022
24 Nov 23, 2022
4.7k Jan 08, 2023
5 Feb 18, 2022
1 Nov 03, 2021
61 Jan 06, 2023
3 Nov 03, 2022
5 Oct 20, 2022