DrugnomeAI: gene druggability

• Introduction
• Citation
• Installation
• Run
  • drugnomeai

Introduction

DrugnomeAI is an adapatation of mantis-ml that provides both disease-agnostic and disease-specific gene druggability framework, implementing stochastic semi-supervised learning on top of scikit-learn and keras/tensorflow.
Drugnome takes its name from a contraction of the druggable genome.

Citation

DrugnomeAI is an ensemble machine-learning framework for predicting druggability of candidate drug targets
Arwa Raies, Ewa Tulodziecka, James Stainer, Lawrence Middleton, Ryan S. Dhindsa, Pamela Hill, Ola Engkvist, Andrew R. Harper, Slavé Petrovski, Dimitrios Vitsios^†
Communications Biology Vol. 5, 1291 (2022)

Installation

Requirements: Python3 (tested with v3.6.7)

git clone https://github.com/astrazeneca-cgr-publications/DrugnomeAI-release.git
sh global_init.sh
python setup.py install

---

In either case, it is highly recommended to create a new virtual environment (e.g. with conda) before installing DrugnomeAI:

conda create -n drugnome python=3.6
conda config --append channels conda-forge   	# add conda-forge in the channels list
conda activate drugnome			                  # activate the newly created conda environment

---

You may now call the following scripts from the command line:

• drugnomeai: run DrugnomeAI druggability score based on a dummy config file (.yaml)

Run each command with -h to see all available options.

Run

Disease-agnostic analysis

In this mode, the only things required to be specified are output directory and either type of labels or a file with custom seed genes which are as follows:
-o OUTPUT_DIR or --output-dir OUTPUT_DIR and -p clin or --pharos-tag clin
-o OUTPUT_DIR or --output-dir OUTPUT_DIR and -t 1 or --tier-tag 1
-o OUTPUT_DIR or --output-dir OUTPUT_DIR and -k seed_genes_file or --known-genes-file seed_genes_file

disease-agnostic run examples

drugnomeai -o out/Genes_analysis -n 4 -k custom_seed_genes.txt -r pre
drugnomeai -o out/Tier1_run -f -t 1
drugnomeai -o analysis_output -r boruta -p tchem 

---

Disease-specific analysis

You need to provide a config file (.yaml) containing information about the diseases/phenotypes of interest.

Required field:

• Disease/Phenotype terms: terms that characterise a phenotype or disease of interest (free text)

Optional fields:

• Additional associated terms: terms used along with Disease/Phenotype terms to extract additional disease/phenotype-associated features (free text)
• Diseases/Phenotypes to exclude: terms to exclude from disease/phenotype characterisation and feature selection (free text)

Config examples:

# Epilepsy_config.yaml
Disease/Phenotype terms: epileptic, epilepsy, seizure
Additional associated terms: brain, nerve, nervous, neuronal, cerebellum, cerebral, hippocampus, hypothalamus
Diseases/Phenotypes to exclude: 

# CKD_config.yaml
Disease/Phenotype terms: renal, kidney, nephro, glomerul, distal tubule 
Additional associated terms: 
Diseases/Phenotypes to exclude: adrenal

Other example config files can be found under example-input or drugnome_ai/conf.

More about DrugnomeAI

Labels

• There are two types of labels available in DrugnomeAI - Pharos labels and the ones representing Tiers. Only one type can be be used at once, however it is possible to select multiple labels of the same type for a single run.
• The available options are the following:

  • Pharos labels:

    • tclin: Tclin
    • tchem: Tchem
    • tbio: Tbio
    • tdark: Tdark

  • Tiers labels:

    • 1: Tier 1
    • 2: Tier 2
    • 3B: Tier 3A
    • 3A: Tier 3B

Label option may be specified in the following way: -p tchem or -t 3A. For multiple choices, use ' ' (space) as a separator, e.g. -t 1 2 3A etc. In the disease-specific run, labels of a given choice are fetched automatically by DrugnomeAI for the genes being associated with the specified disease.

Supervised learning models

• DrugnomeAI runs 8 different supervised models by default: Extra Trees, Random Forest, SVC, Gradient Boosting, XGBoost, Deep Neural Net, Stacking Classifier and Naive Bayes.
• It is also possible to run DrugnomeAI with the -f / --fast option, which will force DrugnomeAI to train only 4 classifiers: Extra Trees, Random Forest, SVC and Gradient Boosting.
• Additionally, the user may explicitly specify which supervised models to be used for training via the -s option. The available model options are coded as follows:
  • et: Extra Trees
  • rf: Random Forest
  • gb: Gradient Boosting
  • xgb: XGBoost
  • svc: Support Vector Classifier
  • dnn: Deep Neural Net
  • stack: Stacking classifier
  • nb: Naive Bayes

Multiple models may be delimited using a (space) as a separator, e.g. -s et, -s et stack gb etc.

Data sources and features

DrugnomeAI allows to make a choice about features which can be selected from different sets which are the following:

• pharos: denotes all druggability related features including: Pharos, DGIdb, Reactome, InWeb, CTD, STRINGdb
• inter: InterPro features related to protein families, domains and super-families

By default, DrugnomeAI runs with all the features listed above. However, the feature set can be modified with -d option, e.g. -d inter.

The disease-specific analysis, on top of the abovementioned, by default leverages features being specifically related to diseases and they are following: GTEx, Protein Atlas, features derived from collapsing analysis of Human Protein Atlas tissue & RNA expression and MsigDB GO.

The InterPro features can also be selected (choices: dom, fam, sup) by user with the option -x dom fam or --inter-pro dom fam - note that the options are delimited with space.

There is also an option to extend generic DrugnomeAI features with -m or --mantis-ml argument. The option --mantis-ml denotes attributes derived from mantis-ml which includes the following: ExAC features, Essential Mouse Genes ones, GnomAD, Genic Intolerance Scores, GWAS & MGI Essential features.

Another available option is -l or --genic-intol which allows you to include solely Genic Intolerance Scores out of all the mantis-ml derived features.

drugnomeai

You need to provide an output directory and label choice or these along with a config file (.yaml).
You may also:

• run specific modules of DrugnomeAI (option -r, default: all) - the available options are:
  • all: all modules are run, meaning end-to-end analysis shall be performed
  • pre: exploratory data analysis shall be run incl. data cleaning & visualizations
  • boruta: feature selection with Boruta algorithms on the already prepared data
  • pu: semi-supervised learning run
  • post: postprocessing of the results yielded by semi-supervised learning
  • post_unsup: dimensionality reduction & clustering run on the results
  • debug: initial exploratory analysis and 1 iteration of semi-supervised learning will be run without Inweb and StringDB features creation
• choose what tier of druggability you would like to train on (-t option; default: 1,2,3A,3B)
• choose to train on pharos based labels (-p option; default: clin,chem,bio,dark)
• define the number of threads to use (-n option; default value: 4).
• define the number of stochastic iterations (-i option; default value: 10)
• provide a file with custom seed genes (-k option; file should contain new-line separated HGNC names without header; bypasses HPO)

disease-agnostic run

drugnomeai -o [output_dir] [-n nthreads] [-i iterations] -k [custom_seed_genes.txt] [-r pre]
or 
drugnomeai -o [output_dir] [-n nthreads] [-i iterations] [-r pre] -t [druggability tier]
or
drugnomeai -o [output_dir] [-n nthreads] [-i iterations] [-r pre] -p [pharos druggability label]

disease-specific run

drugnomeai -c [config_file] -o [output_dir] [-n nthreads] [-i iterations] [-k custom_seed_genes.txt] [-r pre]
or 
drugnomeai -c [config_file] -o [output_dir] [-n nthreads] [-i iterations] [-r pre] [-p pharos druggability label]

Example

drugnomeai -c CKD_config.yaml -o ./CKD-run -f -p tclin
drugnomeai -o ./CKD-run -f -k kcd_seed_genes.txt
drugnomeai -c Epilepsy_config.yaml -o /tmp/Epilepsy-testing -n 20 -r boruta -p tclin

drugnomeai Output

DrugnomeAI predictions for all genes and across all classifiers can be found at [output_dir]/Gene-Predictions.
The AUC_performance_by_Classifier.pdf file under the same dir contains information about the AUC performance per classifier and thus informs about the best performing classifier.

Output figures from all steps during the drugnomeai run (e.g. Exploratory Data Analysis/EDA, supervised-learning, unsupervised-learning) can be found under [output_dir]/Output-Figures.