Figure 2
Flow chart of the model training and prediction process.
We collected 1,256 drug molecules and 611 ligand-bindable targets (a) to constructed an in silico chemical-protein interactome (CPI) using docking (b). Based on the existing drug-indication knowledge, machine learning models (c) were trained to predict drug indications (d) based on the CPI. When a user submits a molecule to our server (e), it is docked against our library targets to generate docking scores. These scores are fed to the machine learning models (f) to predict the indications (g) for this molecule.
