Limitations-compound-potency-predictions

Repository for the paper: Rationalizing principal limitations in assessing and comparing methods for compound potency prediction

Environment

Anaconda can be used to install the .yml file provided.

Notebooks

• regression_models_50_50.ipynb - generates regression models with a 50/50% TR/TE split.
• regression_models_3_bins_330.ipynb - generates regression models with balanced training sets of increasing size, as well as, test sets.
• regression_models_3_bins_330_unbalanced.ipynb - generates regression models with unbalanced training sets of increasing size and test sets.
• result_analysis.ipynb - derives analysis for the computed results.
• Notebooks with R2 in the name correspond to models trained with R2 as a cost function instead of MAE.

Folders

• ML - contains Python scripts to support model building and data analysis.
• dataset - contains the dataset used in the analysis
• figures - contains the manuscript figures