drexml 1.1.0

(DRExM³L) Drug REpurposing using and eXplainable Machine Learning and Mechanistic Models of signal transduction"

Drug REpurposing using eXplainable Machine Learning and Mechanistic Models of signal transduction

Repository for the drexml python package: (DRExM³L) Drug REpurposing using eXplainable Machine Learning and Mechanistic Models of signal transduction

Setup

To install the drexml package use the following:

conda create -n drexml python=3.10
conda activate drexml
pip install drexml

If a CUDA~10.2/11.x (< 12) compatible device is available use:

conda create -n drexml --override-channels -c "nvidia/label/cuda-11.8.0" -c conda-forge cuda cuda-nvcc cuda-toolkit gxx=11.2 python=3.10
conda activate drexml
pip install --no-cache-dir --no-binary=shap drexml

To install drexml in an existing environment, activate it and use:

pip install drexml

Note that by default the setup will try to compile the CUDA modules, if not possible it will use the CPU modules.

Run

To run the program for a disease map that uses circuits from the preprocessed KEGG pathways and the KDT standard list, construct an environment file (e.g. disease.env):

• using the following template if you have a set of seed genes (comma-separated):

seed_genes=2175,2176,2189

• using the following template if you want to use the DisGeNET [1] curated gene-disease associations as seeds.

disease_id="C0015625"

• using the following template if you know which circuits to include (the disease map):

circuits=circuits.tsv.gz

The TSV file circuits.tsv has the following format (tab delimited):

index	in_disease
P-hsa03320-37	0
P-hsa03320-61	0
P-hsa03320-46	0
P-hsa03320-57	0
P-hsa03320-64	0
P-hsa03320-47	0
P-hsa03320-65	0
P-hsa03320-55	0
P-hsa03320-56	0
P-hsa03320-33	0
P-hsa03320-58	0
P-hsa03320-59	0
P-hsa03320-63	0
P-hsa03320-44	0
P-hsa03320-36	0
P-hsa03320-30	0
P-hsa03320-28	1

where:

• index: Hipathia circuit id
• in_disease: (boolean) True/1 if a given circuit is part of the disease

Note that in all cases you can restrict the circuits to the physiological list by setting use_physio=true in the env file.

To run the experiment using 10 CPU cores and 0 GPUs, run the following command within an activated environment:

drexml run --n-gpus 0 --n-cpus 10 $DISEASE_PATH

where:

• --n-gpus indicates the number of gpu devices to use in parallel (-1 -> all) (0 -> None)
• --n-cpus indicates the number of cpu devices to use in parallel (-1 -> all) 8
• DISEASE_PATH indicates the path to the disease env file (e.g. /path/to/disease/folder/disease.env)

Use the --debug option for testing that everything works using a few iterations.

Note that the first time that the full program is run, it will take longer as it downloads the latest versions of each background dataset from Zenodo:

https://doi.org/10.5281/zenodo.6020480

Contribute to development

The recommended setup is:

• setup pipx
• setup miniforge
• use pipx to install pdm
• ensure that pdm is version >=2.1, otherwise update with pipx
• use pipx to inject pdm-bump into pdm
• use pipx to install nox
• run pdm config venv.backend conda
• run make, if you want to use a CUDA enabled GPU, use make gpu=1
• (Recommended): For GPU development, clear the cache using pdm clean cache first

Documentation

The documentation can be found here:

https://loucerac.github.io/drexml/

References

[1] Janet Piñero, Juan Manuel Ramírez-Anguita, Josep Saüch-Pitarch, Francesco Ronzano, Emilio Centeno, Ferran Sanz, Laura I Furlong. The DisGeNET knowledge platform for disease genomics: 2019 update. Nucl. Acids Res. (2019) doi:10.1093/nar/gkz1021