This miniWebinar presents an overview of OpenEye's first release of the package 'Machine Learning Model Building' in Orion. Recent years have seen a massive surge in the use of artificial intelligence (AI) in drug design. However, AI-based predictions have been criticized for being an "unknown box" and frequently fail to convince domain experts.
OpenEye's Machine Learning package trains multiple neural network models to predict the properties of small molecules. We provide a comprehensive report to analyze and choose the best model for future prediction of unseen molecules. In addition, built models provide confidence in prediction and molecular explanation so that domain experts can better understand and evaluate the predictions.
As an exemplary use of the model building package, the release includes a floe which predicts the solubility of small molecules in log(uM). In this webinar, Dr. Mandal will demonstrate the utilities of the package, including:
Introduction and Background
Features of the Solubility prediction Floe
Model Agnostic Molecule Explainer
Error Bar and Domain of Application Analysis of Prediction