ersilia-os/h3d-screening-cascade-code

Code accompanying the manuscript on the first fully automated AI/ML virtual screening cascade implemented at a drug discovery centre in Africa. # h3d-screening-cascade-code Code for the manuscript entitled "First fully automated AI/ML virtual screening cascade implemented at a drug discovery centre in Africa" An outline of the code in each python file is provided below: * **auroc_improvement.py:** Find the change in AUROC between two models relative to a perfect model (AUROC = 1) and the associated standard error. * **binarizer.py:** Classify experimental data points _inclusive of assay cutoff values_ into a binary label, according to whether a large or small experimental readout is the 'active' outcome in that assay. * **cross_val.py:** Find the mean roc curve for a given number of data folds, as well as the median auroc score and standard error. * **ecfp_descriptor.py:** Featurise molecule SMILES into circular Morgan Fingerprints 2048 length with a radius of 3. * **hit_enrichment.py:** Calculate the additional active compounds found by using model predictions after having tested different fractions of the dataset. * **model_performance.py:** Traditional SKlearn ML performance metrics. * **projections.py:** Dimensionality reduction methods (PCA and UMAP) to project high-dimensional molecule descriptions onto two-dimensional space for visualisation.