This repo contains the data and code needed to reproduce the methodology and results for the paper "Deconvoluting Low Yield from Weak Potency in Direct-to-Biology Workflows with Machine Learning".
The necessary packages needed can be install by running pip install . in the directory of the repo.
The folder data contains the raw experimental data from both the initial direct-to-biology screen as well as the subsequent prospective screens.
The folder scripts contains the python scripts used for training and inferencing the machine learning models used.
The folder notebooks contains notebooks for generating the figures and calculating the metrics in the paper, as well as the cheminformatics code used to enumerate building blocks from EnamineREAL.