Abstract:We investigate the impact of choosing regressors and molecular representations for the construction of fast machine learning (ML) models of thirteen electronic ground-state properties of organic molecules. The performance of each regressor/representation/property combination is assessed using learning curves which report out-of-sample errors as a function of training set size with up to $\sim$117k distinct molecules. Molecular structures and properties at hybrid density functional theory (DFT) level of theory used for training and testing come from the QM9 database [Ramakrishnan et al, {\em Scientific Data} {\bf 1} 140022 (2014)] and include dipole moment, polarizability, HOMO/LUMO energies and gap, electronic spatial extent, zero point vibrational energy, enthalpies and free energies of atomization, heat capacity and the highest fundamental vibrational frequency. Various representations from the literature have been studied (Coulomb matrix, bag of bonds, BAML and ECFP4, molecular graphs (MG)), as well as newly developed distribution based variants including histograms of distances (HD), and angles (HDA/MARAD), and dihedrals (HDAD). Regressors include linear models (Bayesian ridge regression (BR) and linear regression with elastic net regularization (EN)), random forest (RF), kernel ridge regression (KRR) and two types of neural net works, graph convolutions (GC) and gated graph networks (GG). We present numerical evidence that ML model predictions deviate from DFT less than DFT deviates from experiment for all properties. Furthermore, our out-of-sample prediction errors with respect to hybrid DFT reference are on par with, or close to, chemical accuracy. Our findings suggest that ML models could be more accurate than hybrid DFT if explicitly electron correlated quantum (or experimental) data was available.

Machine Learning Deciphered Molecular Mechanistics with Accurate Kinetic and Thermodynamic Prediction

Machine Learning in Molecular Simulations of Biomolecules

A Quantum Chemical Convolutional Neural Network Model for Predicting Thermodynamics and Kinetics of DNA Molecules from Sequences

Machine Learning in Molecular Dynamics Simulations of Biomolecular Systems

Modern Semiempirical Electronic Structure Methods and Machine Learning Potentials for Drug Discovery: Conformers, Tautomers, and Protonation States

Direct Learning Hidden Excited State Interaction Patterns from Ab Initio Dynamics and Its Implication As Alternative Molecular Mechanism Models

Machine learning for molecular simulation

A physics-inspired approach to the understanding of molecular representations and models

Machine learning model for non-equilibrium structures and energies of simple molecules

Can molecular dynamics simulations improve predictions of protein-ligand binding affinity with machine learning?

Prediction of transition state structures of gas-phase chemical reactions via machine learning

Prediction Errors of Molecular Machine Learning Models Lower Than Hybrid DFT Error.

Machine Learning Frenkel Hamiltonian Parameters to Accelerate Simulations of Exciton Dynamics

Machine Learning for Molecular Thermodynamics

Machine Learning of Molecular Electronic Properties in Chemical Compound Space

Deciphering Pathways and Thermodynamics of Protein Assembly Using Native Mass Spectrometry

Machine learning for transition state geometry prediction

Predicting Molecular Trajectories using Machine Learning Methods

Nonadiabatic Excited-State Dynamics with Machine Learning.

Elucidating molecular mechanisms of functional conformational changes of proteins via Markov state models