Multitask methods for predicting molecular properties from heterogeneous data

Abstract

Data generation remains a bottleneck in training surrogate models to predict molecular properties. We demonstrate that multitask Gaussian process regression overcomes this limitation by leveraging both expensive and cheap data sources. In particular, we consider training sets constructed from coupled-cluster (CC) and density functional theory (DFT) data. We report that multitask surrogates can predict at CC-level accuracy with a reduction to data generation cost by over an order of magnitude. Of note, our approach allows the training set to include DFT data generated by a heterogeneous mix of exchange-correlation functionals without imposing any artificial hierarchy on functional accuracy. More generally, the multitask framework can accommodate a wider range of training set structures -- including full disparity between the different levels of fidelity -- than existing kernel approaches based on $\Delta$-learning, though we show that the accuracy of the two approaches can be similar. Consequently, multitask regression can be a tool for reducing data generation costs even further by opportunistically exploiting existing data sources.

arXiv comment21 pages, 9 figures

References in corpus (22)

• Gaussian Approximation Potentials: the accuracy of quantum mechanics, without the electrons
• On representing chemical environments
• AFLOW: An automatic framework for high-throughput materials discovery
• Big Data meets Quantum Chemistry Approximations: The $\Delta$-Machine Learning Approach
• DScribe: Library of Descriptors for Machine Learning in Materials Science
• Comparing molecules and solids across structural and alchemical space
• The Open Catalyst 2020 (OC20) Dataset and Community Challenges
• Perspective: How good is DFT for water?
• Physics-inspired structural representations for molecules and materials
• Ab initio theory and modeling of water
• ANI-1: A data set of 20M off-equilibrium DFT calculations for organic molecules
• AiiDA 1.0, a scalable computational infrastructure for automated reproducible workflows and data provenance
• Interactive Supercomputing on 40,000 Cores for Machine Learning and Data Analysis
• q-AQUA: a many-body CCSD(T) water potential, including 4-body interactions, demonstrates the quantum nature of water from clusters to the liquid phase
• Thoughts on Massively Scalable Gaussian Processes
• Transfer learning for chemically accurate interatomic neural network potentials
• Multi-Fidelity Machine Learning for Excited State Energies of Molecules
• Optimized Multifidelity Machine Learning for Quantum Chemistry
• Black-box inhomogeneous preconditioning for self-consistent field iterations in density functional theory
• A robust and efficient line search for self-consistent field iterations
• PAC-Bayes Compression Bounds So Tight That They Can Explain Generalization
• Numerical stability and efficiency of response property calculations in density functional theory

Cited by in corpus (3)

• Investigating Data Hierarchies in Multifidelity Machine Learning for Excitation Energies
• Assessing Non-Nested Configurations of Multifidelity Machine Learning for Quantum-Chemical Properties
• Data Fusion of Deep Learned Molecular Embeddings for Property Prediction