Drug2cell

This is a collection of utility functions for gene group activity evaluation in scanpy, both for per-cell scoring and marker-based enrichment/overrepresentation analyses. Drug2cell makes use of established methodology, and offers it in a convenient/efficient form for easy scanpy use. The package comes with a set of ChEMBL derived drug target sets, but has straightforward input formatting so it can be easily used with any gene groups of choice.

Installation

Once public, this will be on pip. For now install straight from this repository:

pip install git+https://github.com/Teichlab/drug2cell.git

Drug2cell's GSEA makes use of blitzGSEA, which needs to be installed from GitHub:

pip install git+https://github.com/MaayanLab/blitzgsea.git

Usage

import drug2cell as d2c

Per-cell scoring can be done via d2c.score(), and creates a new gene group feature space object in .uns['drug2cell'] of the supplied object. The .obs and .obsm of the original object are copied over for ease of downstream use, like plotting or potential marker detection.

Enrichment is done with GSEA, contained within the function d2c.gsea(). Overrepresentation analysis can be done with the hypergeometric test of d2c.hypergeometric(). Both of those require having ran sc.tl.rank_genes_groups() on the input object, and return one data frame per evaluated cluster with enrichment/overrepresentation results.

It's possible to provide your own gene groups as a dictionary, with the names of the groups as keys and the corresponding gene lists as the values. Pass this dictionary as the targets argument of any of those three functions.

Please refer to the demo notebook, all of the input arguments are detailed in ReadTheDocs.

ChEMBL parsing

Drug2cell also features instructions on how to parse the ChEMBL database into drugs and their targets. Some additional notebooks are included. Both refer to a pre-parsed data frame of ChEMBL human targets, which can currently be accessed at /nfs/team205/kk18/data/ChEMBL/pkls/chembl_30_merged_genesymbols_humans.pkl. Once public, this will be put up on FTP.

• Filtering shows how this data frame was turned into the drugs:targets dictionary shipped with the package by default. There are some helper functions included in d2c.chembl which can assist you shall you wish to filter it in a different way.
• Initial database parsing shows how the data frame was created from online resources.

Citation

Kazumasa Kanemaru and James Cranley et al., Spatially resolved multiomics of human cardiac niches. bioRxiv 2023.01.30.526202. Link