Learning to Make Chemical Predictions: the Interplay of Feature Representation, Data, and Machine Learning Algorithms

Abstract

Recently supervised machine learning has been ascending in providing new predictive approaches for chemical, biological and materials sciences applications. In this Perspective we focus on the interplay of machine learning algorithm with the chemically motivated descriptors and the size and type of data sets needed for molecular property prediction. Using Nuclear Magnetic Resonance chemical shift prediction as an example, we demonstrate that success is predicated on the choice of feature extracted or real-space representations of chemical structures, whether the molecular property data is abundant and/or experimentally or computationally derived, and how these together will influence the correct choice of popular machine learning algorithms drawn from deep learning, random forests, or kernel methods.

References in corpus (10)

• Adam: A Method for Stochastic Optimization
• On representing chemical environments
• SchNet - a deep learning architecture for molecules and materials
• All-electron magnetic response with pseudopotentials: NMR chemical shifts
• Explainable Machine Learning for Scientific Insights and Discoveries
• Machine learning prediction errors better than DFT accuracy
• sGDML: Constructing Accurate and Data Efficient Molecular Force Fields Using Machine Learning
• Transferability in Machine Learning for Electronic Structure via the Molecular Orbital Basis
• On the Completeness of Atomic Structure Representations
• Training neural nets to learn reactive potential energy surfaces using interactive quantum chemistry in virtual reality