Machine learning for molecular properties is an area of machine learning that has seen massive innovation in the past decade. Learning about these topics is important for their impact on materials engineering and also for understanding how knowledge of the physics underlying a problem influence the design of machine learning algorithms.
The major goals of this module include:
- Highlighting roots of modern molecular machine learning in chemoinformatics. What is QSAR and when should I use it?
- Identifying an appropriate class of machine learning models for molecules. When would you use graph convolution network for predicting molecular toxicity?
- Explaining why (nearly) all neural networks for molecular properties are "message passing neural networks." What sets MEGNet and Gilmer's MPNN apart?
- Describing the key features, major methods and disadvantages of kernel-based machine learning. What are the key features of FCHL?
- Presenting history of machine learning for molecules. How are SchNet and DTNN related?