Resolving transition metal chemical space: feature selection for machine learning and structure-property relationships

Abstract

Machine learning (ML) of quantum mechanical properties shows promise for accelerating chemical discovery. For transition metal chemistry where accurate calculations are computationally costly and available training data sets are small, the molecular representation becomes a critical ingredient in ML model predictive accuracy. We introduce a series of revised autocorrelation functions (RACs) that encode relationships between the heuristic atomic properties (e.g., size, connectivity, and electronegativity) on a molecular graph. We alter the starting point, scope, and nature of the quantities evaluated in standard ACs to make these RACs amenable to inorganic chemistry. On an organic molecule set, we first demonstrate superior standard AC performance to other presently-available topological descriptors for ML model training, with mean unsigned errors (MUEs) for atomization energies on set-aside test molecules as low as 6 kcal/mol. For inorganic chemistry, our RACs yield 1 kcal/mol ML MUEs on set-aside test molecules in spin-state splitting in comparison to 15-20x higher errors from feature sets that encode whole-molecule structural information. Systematic feature selection methods including univariate filtering, recursive feature elimination, and direct optimization (e.g., random forest and LASSO) are compared. Random-forest- or LASSO-selected subsets 4-5x smaller than RAC-155 produce sub- to 1-kcal/mol spin-splitting MUEs, with good transferability to metal-ligand bond length prediction (0.004-5 Å MUE) and redox potential on a smaller data set (0.2-0.3 eV MUE). Evaluation of feature selection results across property sets reveals the relative importance of local, electronic descriptors (e.g., electronegativity, atomic number) in spin-splitting and distal, steric effects in redox potential and bond lengths.

43 double spaced pages, 11 figures, 4 tables

References in corpus (25)

• On representing chemical environments
• Fast and Accurate Modeling of Molecular Atomization Energies with Machine Learning
• ANI-1: An extensible neural network potential with DFT accuracy at force field computational cost
• Big Data of Materials Science - Critical Role of the Descriptor
• Big Data meets Quantum Chemistry Approximations: The $\Delta$-Machine Learning Approach
• Comparing molecules and solids across structural and alchemical space
• Machine learning prediction errors better than DFT accuracy
• Finding Density Functionals with Machine Learning
• Low Hole Effective Mass p-type Transparent Conducting Oxides: Identification and Design Principles
• How to represent crystal structures for machine learning: towards fast prediction of electronic properties
• Prediction model of band-gap for AX binary compounds by combination of density functional theory calculations and machine learning techniques
• Understanding molecular representations in machine learning: The role of uniqueness and target similarity
• Predicting Electronic Structure Properties of Transition Metal Complexes with Neural Networks
• Deep learning and the Schrödinger equation
• Accelerated materials property predictions and design using motif-based fingerprints
• Quantum Chemistry for Solvated Molecules on Graphical Processing Units (GPUs)using Polarizable Continuum Models
• Orbital-free Bond Breaking via Machine Learning
• Machine Learning Exciton Dynamics
• Assessment of density functional theory for iron(II) molecules across the spin-crossover transition
• The Many-Body Expansion Combined with Neural Networks
• Towards quantifying the role of exact exchange in predictions of transition metal complex properties
• Where Does the Density Localize? Convergent Behavior for Global Hybrids, Range Separation, and DFT+U
• Gradient-Driven Molecule Construction: An Inverse Approach Applied to the Design of Small-Molecule Fixating Catalysts
• Fast Cross-Validation via Sequential Testing
• Constant Size Molecular Descriptors For Use With Machine Learning

Cited by in corpus (19)

• Big-Data Science in Porous Materials: Materials Genomics and Machine Learning
• WACSF - Weighted Atom-Centered Symmetry Functions as Descriptors in Machine Learning Potentials
• Autonomous discovery in the chemical sciences part II: Outlook
• Computer-Aided Multi-Objective Optimization in Small Molecule Discovery
• Maximizing information from chemical engineering data sets: Applications to machine learning
• Error-Controlled Exploration of Chemical Reaction Networks with Gaussian Processes
• Metadynamics for Training Neural Network Model Chemistries: a Competitive Assessment
• Putting Density Functional Theory to the Test in Machine-Learning-Accelerated Materials Discovery
• Transfer learning for chemically accurate interatomic neural network potentials
• CRAFTED -- An exploratory database of simulated adsorption isotherms of metal-organic frameworks
• Towards Linearly Scaling and Chemically Accurate Global Machine Learning Force Fields for Large Molecules
• Predicting Oxidation and Spin States by High-Dimensional Neural Networks: Applications to Lithium Manganese Oxide Spinels
• An Atomistic Machine Learning Package for Surface Science and Catalysis
• Representations and Strategies for Transferable Machine Learning Models in Chemical Discovery
• Machine Learning Based Prediction of Proton Conductivity in Metal-Organic Frameworks
• Quantum chemical roots of machine-learning molecular similarity descriptors
• Ab initio machine learning in chemical compound space
• Prediction of Atomization Energies of Au13+ Clusters through the Machine Learning Approach
• Bond type restricted radial distribution functions for accurate machine learning prediction of atomization energies