Abstract: Chemprop is an open-source implementation of a directed message passing neural network (D-MPNN) that has been demonstrated to be successful in predicting a variety of molecular properties, including solvation properties, optical properties, infrared spectra, and toxicity. The framework has also been used to predict reaction properties and to identify new antibiotics. Chemprop can be used via a command-line interface or within a Python script or Jupyter notebook. In this tutorial, we will discuss the D-MPNN algorithm and several successful use cases, followed by a demonstration of some of Chemprop’s core functionalities (e.g. training and testing models for single- and multiple-molecule properties and reactions, transfer learning, and estimating uncertainty).

Series information: Our series is aimed at active researchers and educators and designed to introduce practical skills with online, hands-on activities that participants will be able to incorporate in their own work. Hands-on activities will use nanoHUB cloud computing resources, negating the need to download or install any software. All that is required of the audience is an internet connection and an hour to spare for the demonstration. After the training sessions, you will be able to continue using nanoHUB for research or education.

Bio: Kevin Greenman is a Ph.D. candidate in chemical engineering and computation at MIT working under the supervision of Profs. Rafael Gómez-Bombarelli and William Green. During his undergraduate studies at the University of Michigan, he studied the structural, thermodynamic, and optical properties of nitride semiconductors, and he developed a nanoHUB tool for research and education on computational catalysis during a summer REU at Purdue University. His current research interests are in using the combination of atomistic simulations and machine learning with experimental collaboration for the discovery and design of new molecules and materials.
 

1. data science
2. machine learning
3. neural networks