Less is more: sampling chemical space with active learning

Abstract

The development of accurate and transferable machine learning (ML) potentials for predicting molecular energetics is a challenging task. The process of data generation to train such ML potentials is a task neither well understood nor researched in detail. In this work, we present a fully automated approach for the generation of datasets with the intent of training universal ML potentials. It is based on the concept of active learning (AL) via Query by Committee (QBC), which uses the disagreement between an ensemble of ML potentials to infer the reliability of the ensemble's prediction. QBC allows the presented AL algorithm to automatically sample regions of chemical space where the ML potential fails to accurately predict the potential energy. AL improves the overall fitness of ANAKIN-ME (ANI) deep learning potentials in rigorous test cases by mitigating human biases in deciding what new training data to use. AL also reduces the training set size to a fraction of the data required when using naive random sampling techniques. To provide validation of our AL approach we develop the COMP6 benchmark (publicly available on GitHub), which contains a diverse set of organic molecules. Through the AL process, it is shown that the AL-based potentials perform as well as the ANI-1 potential on COMP6 with only 10% of the data, and vastly outperforms ANI-1 with 25% the amount of data. Finally, we show that our proposed AL technique develops a universal ANI potential (ANI-1x) that provides accurate energy and force predictions on the entire COMP6 benchmark. This universal ML potential achieves a level of accuracy on par with the best ML potentials for single molecule or materials, while remaining applicable to the general class of organic molecules comprised of the elements CHNO.

Accepted at J. Chem. Phys

References in corpus (6)

• ANI-1: An extensible neural network potential with DFT accuracy at force field computational cost
• ANI-1: A data set of 20M off-equilibrium DFT calculations for organic molecules
• Hierarchical modeling of molecular energies using a deep neural network
• Structure-based Sampling and Self-correcting Machine Learning for Accurate Calculations of Potential Energy Surfaces and Vibrational Levels
• Metadynamics for Training Neural Network Model Chemistries: a Competitive Assessment
• Parallel and Distributed Thompson Sampling for Large-scale Accelerated Exploration of Chemical Space

Cited by in corpus (147)

• Machine learning and the physical sciences
• Machine Learning Force Fields
• DP-GEN: A concurrent learning platform for the generation of reliable deep learning based potential energy models
• Active Learning of Uniformly Accurate Inter-atomic Potentials for Materials Simulation
• Opportunities and Challenges for Machine Learning in Materials Science
• GPUMD: A package for constructing accurate machine-learned potentials and performing highly efficient atomistic simulations
• Machine learning for electronically excited states of molecules
• Autonomous discovery in the chemical sciences part I: Progress
• Towards self-driving laboratories: The central role of density functional theory in the AI age
• Machine learning enables long time scale molecular photodynamics simulations
• A practical guide to machine learning interatomic potentials -- Status and future
• Coarse Graining Molecular Dynamics with Graph Neural Networks
• Evaluation of the MACE Force Field Architecture: from Medicinal Chemistry to Materials Science
• Committee neural network potentials control generalization errors and enable active learning
• Efficient global structure optimization with a machine learned surrogate model
• Active learning of reactive Bayesian force fields: Application to heterogeneous hydrogen-platinum catalysis dynamics
• Computer-Aided Multi-Objective Optimization in Small Molecule Discovery
• DPA-2: a large atomic model as a multi-task learner
• On the redundancy in large material datasets: efficient and robust learning with less data
• End-to-End Differentiable Molecular Mechanics Force Field Construction
• Guest Editorial: Special Topic on Data-enabled Theoretical Chemistry
• Uncertainty estimation for molecular dynamics and sampling
• Automated discovery of a robust interatomic potential for aluminum
• Strategies for the Construction of Machine-Learning Potentials for Accurate and Efficient Atomic-Scale Simulations
• DeePKS: a comprehensive data-driven approach towards chemically accurate density functional theory
• Tutorial: How to Train a Neural Network Potential
• Ground state energy functional with Hartree-Fock efficiency and chemical accuracy
• The Materials Simulation Toolkit for Machine Learning (MAST-ML): an automated open source toolkit to accelerate data-driven materials research
• Global optimization of atomistic structure enhanced by machine learning
• MLatom 3: Platform for machine learning-enhanced computational chemistry simulations and workflows
• Active Training of Physics-Informed Neural Networks to Aggregate and Interpolate Parametric Solutions to the Navier-Stokes Equations
• Machine learning and excited-state molecular dynamics
• Gaussian Process-Based Refinement of Dispersion Corrections
• Universal QM/MM Approaches for General Nanoscale Applications
• Ensemble Learning of Coarse-Grained Molecular Dynamics Force Fields with a Kernel Approach
• Bayesian Force Fields from Active Learning for Simulation of Inter-Dimensional Transformation of Stanene
• TorchMD-Net 2.0: Fast Neural Network Potentials for Molecular Simulations
• Deep Ensembles vs. Committees for Uncertainty Estimation in Neural-Network Force Fields: Comparison and Application to Active Learning
• An Experimentally Driven Automated Machine Learned lnter-Atomic Potential for a Refractory Oxide
• Atomic Permutationally Invariant Polynomials for Fitting Molecular Force Fields
• Predicting tensorial molecular properties with equivariant machine-learning models
• Accurate machine learning force fields via experimental and simulation data fusion
• Deep Coarse-grained Potentials via Relative Entropy Minimization
• Machine-learned molecular mechanics force field for the simulation of protein-ligand systems and beyond
• Scalable Hybrid Deep Neural Networks/Polarizable Potentials Biomolecular Simulations including long-range effects
• Transfer learning for chemically accurate interatomic neural network potentials
• Active learning of the thermodynamics-dynamics tradeoff in protein condensates
• Lifelong Machine Learning Potentials
• Developments and Further Applications of Ephemeral Data Derived Potentials
• Predicting Properties of Periodic Systems from Cluster Data: A Case Study of Liquid Water
• Fast and Sample-Efficient Interatomic Neural Network Potentials for Molecules and Materials Based on Gaussian Moments
• Machine learning for molecular dynamics with strongly correlated electrons
• Elinvar effect in $\beta -$Ti simulated by on-the-fly trained moment tensor potential
• Machine Learned Hückel Theory: Interfacing Physics and Deep Neural Networks
• Modeling nanoconfinement effects using active learning
• A first-principles machine-learning force field for heterogeneous ice nucleation on microcline feldspar
• Thermophysical properties of FLiBe using moment tensor potentials
• Machine Learning Prediction of DNA Charge Transport
• Deep Neural Network Computes Electron Densities and Energies of a Large Set of Organic Molecules Faster than Density Functional Theory (DFT)
• Atomistic Structure Learning Algorithm with surrogate energy model relaxation
• E(n)-Equivariant Cartesian Tensor Passing Potential
• On the design space between molecular mechanics and machine learning force fields
• Active Learning A Neural Network Model For Gold Clusters \& Bulk From Sparse First Principles Training Data
• ParAMS: Parameter Optimization for Atomistic and Molecular Simulations
• Machine Learning and Data-Driven Methods in Computational Surface and Interface Science
• Integrating Machine Learning with Physics-Based Modeling
• Lightweight and Effective Tensor Sensitivity for Atomistic Neural Networks
• Denoise Pretraining on Nonequilibrium Molecules for Accurate and Transferable Neural Potentials
• Impact of the characteristics of quantum chemical databases on machine learning predictions of tautomerization energies
• Deep Ensemble Bayesian Active Learning : Addressing the Mode Collapse issue in Monte Carlo dropout via Ensembles
• Efficient Estimation of Material Property Curves and Surfaces via Active Learning
• High-dimensional neural network potentials for accurate vibrational frequencies: The formic acid dimer benchmark
• Enhancing the Accuracy of Density Functional Tight Binding Models Through ChIMES Many-body Interaction Potentials
• Machine Learned Interatomic Potential for Dispersion Strengthened Plasma Facing Components
• Accelerated Organic Crystal Structure Prediction with Genetic Algorithms and Machine Learning
• In operando active learning of interatomic interaction during large-scale simulations
• 3DReact: Geometric deep learning for chemical reactions
• Prediction uncertainty validation for computational chemists
• SCINE -- Software for Chemical Interaction Networks
• Deep Learning to Estimate Permeability using Geophysical Data
• Machine Learned Interatomic Potentials for Ternary Carbides trained on the AFLOW Database
• Ultra-Fast Semi-Empirical Quantum Chemistry for High-Throughput Computational Campaigns with Sparrow
• Active Learning Algorithm for Computational Physics
• Machine Learning Diffusion Monte Carlo Energies
• Ring Polymer Molecular Dynamics and Active Learning of Moment Tensor Potential for Gas-Phase Barrierless Reactions: Application to S + H2
• Reducing the cost of neural network potential generation for reactive molecular systems
• Molecular Quantum Chemical Data Sets and Databases for Machine Learning Potentials
• Generator of Neural Network Potential for Molecular Dynamics: Constructing Robust and Accurate Potentials with Active Learning for Nanosecond-scale Simulations
• Deep Learning in Protein Structural Modeling and Design
• A machine learning potential for simulating infrared spectra of nanosilicate clusters
• Random sampling versus active learning algorithms for machine learning potentials of quantum liquid water
• Analytical Gradients for Molecular-Orbital-Based Machine Learning
• Training Algorithm Matters for the Performance of Neural Network Potential: A Case Study of Adam and the Kalman Filter Optimizers
• Mean-Field Density Matrix Decompositions
• Augmentation of Universal Potentials for Broad Applications
• Active Learning Approach to Optimization of Experimental Control
• First principles reactive simulation for equation of state prediction
• Physics-informed active learning for accelerating quantum chemical simulations
• Improved decision making with similarity based machine learning: Applications in chemistry
• Unraveling Antagonistic Collision-Controlled Reactivity in Energetic Molecular Perovskites with Deep Potential Molecular Dynamics
• Facilitating {\it ab initio} configurational sampling of multicomponent solids using an on-lattice neural network model and active learning
• Probing the state of hydrogen in $\delta$-AlOOH at mantle conditions with machine learning potential
• Accurate Molecular Dynamics Enabled by Efficient Physically-Constrained Machine Learning Approaches
• SPA${}^{\mathrm{H}}$M(a,b): encoding the density information from guess Hamiltonian in quantum machine learning representations
• Machine learning potentials for complex aqueous systems made simple
• Discovery of structure-property relations for molecules via hypothesis-driven active learning over the chemical space
• Machine Learning Enhanced Calculation of Quantum-Classical Binding Free Energies
• Farthest Point Sampling in Property Designated Chemical Feature Space as a General Strategy for Enhancing the Machine Learning Model Performance for Small Scale Chemical Dataset
• Tell machine learning potentials what they are needed for: Simulation-oriented training exemplified for glycine
• Weighted Active Space Protocol for Multireference Machine-Learned Potentials
• Compressing physical properties of atomic species for improving predictive chemistry
• Does Hessian Data Improve the Performance of Machine Learning Potentials?
• Efficient Generation of Stable Linear Machine-Learning Force Fields with Uncertainty-Aware Active Learning
• Evaluating Scalable Uncertainty Estimation Methods for DNN-Based Molecular Property Prediction
• Scaling Graph Neural Networks to Large Proteins
• Molecular dynamics-driven global tetra-atomic potential energy surfaces: Application to the AlF dimer
• Active-learning-based efficient prediction of ab-initio atomic energy: a case study on a Fe random grain boundary model with millions of atoms
• Apax: A Flexible and Performant Framework For The Development of Machine-Learned Interatomic Potentials
• A Planning-and-Exploring Approach to Extreme-Mechanics Force Fields
• Evolutionary Monte Carlo of QM properties in chemical space: Electrolyte design
• Atomistic Simulations of Oxide-Water Interfaces using Machine Learning Potentials
• Accelerating structure search using atomistic graph-based classifiers
• Physical machine learning outperforms "human learning" in Quantum Chemistry
• Free energy profiles for chemical reactions in solution from high-dimensional neural network potentials: The case of the Strecker synthesis
• Lifelong Machine Learning Potentials for Chemical Reaction Network Explorations
• The Enduring Relevance of Semiempirical Quantum Mechanics
• Ab initio machine learning in chemical compound space
• Unveil the unseen: Exploit information hidden in noise
• All-in-one foundational models learning across quantum chemical levels
• Strength of 2D Glasses Explored by Machine-Learning Force Fields
• Turn-Key Constrained Parameter Space Exploration for Particle Accelerators Using Bayesian Active Learning
• Learning atomic forces from uncertainty-calibrated adversarial attacks
• Machine-Learned Potentials for Solvation Modeling
• Understanding ML driven HPC: Applications and Infrastructure
• A Scalable Molecular Force Field Parameterization Method Based on Density Functional Theory and Quantum-Level Machine Learning
• Entropy-based Active Learning of Graph Neural Network Surrogate Models for Materials Properties
• Artificial Intelligence for Direct Prediction of Molecular Dynamics Across Chemical Space
• Incorporating electronic information into Machine Learning potential energy surfaces via approaching the ground-state electronic energy as a function of atom-based electronic populations
• Scalable Training of Neural Network Potentials for Complex Interfaces Through Data Augmentation
• Active learning with physics-informed neural networks for optimal sensor placement in deep tunneling through transversely isotropic elastic rocks
• Simulations of water and hydrophobic hydration using a neural network potential
• Scaffold-Induced Molecular Graph (SIMG): Effective Graph Sampling Methods for High-Throughput Computational Drug Discovery
• Integer linear programming for unsupervised training set selection in molecular machine learning
• Docking-based Virtual Screening with Multi-Task Learning
• Active learning of potential-energy surfaces of weakly-bound complexes with regression-tree ensembles
• Graph Convolutional Neural Networks for (QM)ML/MM Molecular Dynamics Simulations
• Improving Molecular Force Fields Across Configurational Space by Combining Supervised and Unsupervised Machine Learning