ud-crpl / rcdml

Source code for the RCDML workflow. The RCDML workflow can be used to classify rare disease patients as high/low responders for a given drug therapy using RNA-seq count and drug response data.

RCDML

Welcome to the RCDML Project!

The RNA-seq Count Drug-response Machine Learning (RCDML) Workflow is a ML workflow that can be used for drug response classification of rare disease patients. Given drug response data and RNA-seq count data, the model follows the ML workflow below to classify patients as "high responder" or "low responder" for a given inhibitor.

The RCDML pipeline was evaluated using RNA-seq count and drug response data available for Acute Myeloid Leukemia (AML) patients and over 100 different drugs as part of the BeatAML project.

The RCDML source code consist of:

parser.py – Contains the code used to parse the configuration file.

data_preprocessing.py – Contains the code that loads and configures the dataset, and the code for assigning responder/non-responder labels.

feature_selection.py – Contains the implementations of the feature selection techniques and the feature counter generator.

classification.py – Contains the implementation of the classifiers and the hyperparameter optimization technique. The hyperparameter lists used are found here.

validation.py – Contains the code used to create the confusion matrices, ROC curves, run inference and make predictions.

main.py – Contains the framework structure. This is the script that needs to get called to run the ML pipeline.

parameters.cfg – Parameter configuration for the ML pipeline run. The user can select the feature selection techniques, classifiers, and other options that will be used in the run.

/tools/ - Utility tools for gathering results, create feature counters, get family drug list, etc. For more information on each tool follow this wiki page link.

/setup/ - Contains conda environment yml file and parameter configuration presets. For more information on each setup preset follow this wiki page.

For documentation on how to get started with the RCDML workflow visit the wiki.