Using Diffusion Maps to Analyze Reaction Dynamics for a Hydrogen Combustion Benchmark Dataset.

We use local diffusion maps to assess the quality of two types of collective variables (CVs) for a recently published hydrogen combustion benchmark dataset1 that contains ab initio molecular dynamics (MD) trajectories and normal modes along minimum energy paths. This approach was recently advocated in2 for assessing CVs and analyzing reactions modeled by classical MD simulations. We report the effectiveness of this approach to molecular systems modeled by quantum ab initio MD. In addition to assessing the quality of CVs, we also use global diffusion maps to perform committor analysis as proposed in.2 We show that the committor function obtained from the global diffusion map allows us to identify transition regions of interest in several hydrogen combustion reaction channels.

Leak Proof PDBBind: A Reorganized Dataset of Protein-Ligand Complexes for More Generalizable Binding Affinity Prediction.

Many physics-based and machine-learned scoring functions (SFs) used to predict protein-ligand binding free energies have been trained on the PDBBind dataset. However, it is controversial as to whether new SFs are actually improving since the general, refined, and core datasets of PDBBind are cross-contaminated with proteins and ligands with high similarity, and hence they may not perform comparably well in binding prediction of new protein-ligand complexes. In this work we have carefully prepared a cleaned PDBBind data set of non-covalent binders that are split into training, validation, and test datasets to control for data leakage. The resulting leak-proof (LP)-PDBBind data is used to retrain four popular SFs: AutoDock vina, Random Forest (RF)-Score, InteractionGraphNet (IGN), and DeepDTA, to better test their capabilities when applied to new protein-ligand complexes. In particular we have formulated a new independent data set, BDB2020+, by matching high quality binding free energies from BindingDB with co-crystalized ligand-protein complexes from the PDB that have been deposited since 2020. Based on all the benchmark results, the retrained models using LP-PDBBind that rely on 3D information perform consistently among the best, with IGN especially being recommended for scoring and ranking applications for new protein-ligand systems.

M-Chem: a Modular Software Package for Molecular Simulation that Spans Scientific Domains.

We present a new software package called M-Chem that is designed from scratch in C++ and parallelized on shared-memory multi-core architectures to facilitate efficient molecular simulations. Currently, M-Chem is a fast molecular dynamics (MD) engine that supports the evaluation of energies and forces from two-body to many-body all-atom potentials, reactive force fields, coarse-grained models, combined quantum mechanics molecular mechanics (QM/MM) models, and external force drivers from machine learning, augmented by algorithms that are focused on gains in computational simulation times. M-Chem also includes a range of standard simulation capabilities including thermostats, barostats, multi-timestepping, and periodic cells, as well as newer methods such as fast extended Lagrangians and high quality electrostatic potential generation. At present M-Chem is a developer friendly environment in which we encourage new software contributors from diverse fields to build their algorithms, models, and methods in our modular framework. The long-term objective of M-Chem is to create an interdisciplinary platform for computational methods with applications ranging from biomolecular simulations, reactive chemistry, to materials research.

Using machine learning to go beyond potential energy surface benchmarking for chemical reactivity.

We train an equivariant machine learning (ML) model to predict energies and forces for hydrogen combustion under conditions of finite temperature and pressure. This challenging case for reactive chemistry illustrates that ML potential energy surfaces are difficult to make complete, due to overreliance on chemical intuition of what data are important for training. Instead, a 'negative design' data acquisition strategy using metadynamics as part of an active learning workflow helps to create a ML model that avoids unforeseen high-energy or unphysical energy configurations. This strategy more rapidly converges the potential energy surfaces such that it is now more efficient to make calls to the external ab initio source when query-by-committee models disagree to further molecular dynamics in time without need for ML retraining. With the hybrid ML-physics model we realize two orders of magnitude reduction in cost, allowing for prediction of the free-energy change in the transition-state mechanism for several hydrogen combustion reaction channels.

NewtonNet: a Newtonian message passing network for deep learning of interatomic potentials and forces.

We report a new deep learning message passing network that takes inspiration from Newton's equations of motion to learn interatomic potentials and forces. With the advantage of directional information from trainable force vectors, and physics-infused operators that are inspired by Newtonian physics, the entire model remains rotationally equivariant, and many-body interactions are inferred by more interpretable physical features. We test NewtonNet on the prediction of several reactive and non-reactive high quality ab initio data sets including single small molecules, a large set of chemically diverse molecules, and methane and hydrogen combustion reactions, achieving state-of-the-art test performance on energies and forces with far greater data and computational efficiency than other deep learning models.

A benchmark dataset for Hydrogen Combustion.

The generation of reference data for deep learning models is challenging for reactive systems, and more so for combustion reactions due to the extreme conditions that create radical species and alternative spin states during the combustion process. Here, we extend intrinsic reaction coordinate (IRC) calculations with ab initio MD simulations and normal mode displacement calculations to more extensively cover the potential energy surface for 19 reaction channels for hydrogen combustion. A total of ∼290,000 potential energies and ∼1,270,000 nuclear force vectors are evaluated with a high quality range-separated hybrid density functional, ωB97X-V, to construct the reference data set, including transition state ensembles, for the deep learning models to study hydrogen combustion reaction.

Mechanism of the Stereoselective Catalysis of Diels-Alderase PyrE3 Involved in Pyrroindomycin Biosynthesis.

The biosynthesis of pyrroindomycins A and B features a complexity-building [4 + 2] cycloaddition cascade, which generates the spirotetramate core under the catalytic effects of monofunctional Diels-Alderases PyrE3 and PyrI4. We recently showed that the main functions of PyrI4 include acid catalysis and induced-fit/conformational selection. We now present quantum mechanical and molecular dynamics studies implicating a different mode of action by PyrE3, which prearranges an anionic polyene substrate into a high-energy reactive conformation at which an inverse-electron-demand Diels-Alder reaction can occur with a low barrier. Stereoselection is realized by strong binding interactions at the endo stereochemical relationship and a local steric constraint on the endo-1,3-diene unit. These findings, illustrating distinct mechanisms for PyrE3 and PyrI4, highlight how nature has evolved multiple ways to catalyze Diels-Alder reactions.

Protein C-GeM: A Coarse-Grained Electron Model for Fast and Accurate Protein Electrostatics Prediction.

The electrostatic potential (ESP) is a powerful property for understanding and predicting electrostatic charge distributions that drive interactions between molecules. In this study, we compare various charge partitioning schemes including fitted charges, density-based quantum mechanical (QM) partitioning schemes, charge equilibration methods, and our recently introduced coarse-grained electron model, C-GeM, to describe the ESP for protein systems. When benchmarked against high quality density functional theory calculations of the ESP for tripeptides and the crambin protein, we find that the C-GeM model is of comparable accuracy to ab initio charge partitioning methods, but with orders of magnitude improvement in computational efficiency since it does not require either the electron density or the electrostatic potential as input.

Recent Advances for Improving the Accuracy, Transferability, and Efficiency of Reactive Force Fields.

Reactive force fields provide an affordable model for simulating chemical reactions at a fraction of the cost of quantum mechanical approaches. However, classically accounting for chemical reactivity often comes at the expense of accuracy and transferability, while computational cost is still large relative to nonreactive force fields. In this Perspective, we summarize recent efforts for improving the performance of reactive force fields in these three areas with a focus on the ReaxFF theoretical model. To improve accuracy, we describe recent reformulations of charge equilibration schemes to overcome unphysical long-range charge transfer, new ReaxFF models that account for explicit electrons, and corrections for energy conservation issues of the ReaxFF model. To enhance transferability we also highlight new advances to include explicit treatment of electrons in the ReaxFF and hybrid nonreactive/reactive simulations that make it possible to model charge transfer, redox chemistry, and large systems such as reverse micelles within the framework of a reactive force field. To address the computational cost, we review recent work in extended Lagrangian schemes and matrix preconditioners for accelerating the charge equilibration method component of ReaxFF and improvements in its software performance in LAMMPS.

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

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 method 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 methods drawn from deep learning, random forests, or kernel methods.