Themed collection Emerging AI Approaches in Physical Chemistry

This Perspective discusses recent advances in constructing high fidelity diabatic potential energy matrices for nonadiabatic systems and the associated quantum dynamics.

This perspective review features, among others, the use of DFT, QSAR modeling, artificial neural network (ANN) modeling, molecular dynamics simulations and Monte Carlo simulations in modelling organic corrosion inhibitors. It is a compendium of studies on the subject.

Machine learning algorithms can facilitate the reaction prediction in heterogeneous catalysis.

The theoretical description of dissociative chemisorption of isolated molecules on metal surfaces is discussed. Emphasis is put on methods that deliver chemical accuracy for the dissociative chemisorption probability S₀, so that Δ ≤ 1 kcal mol⁻¹.

Phonon-mediated thermal transport is inherently multi-scale. The nature of multi-scale thermal transport is that there are different heat transfer physics across different length scales with strong entanglement and interaction with each other.

High throughput experimentation in heterogeneous catalysis provides an efficient solution to the generation of large datasets under reproducible conditions.

Recently, machine learning (ML) has established itself in various worldwide benchmarking competitions in computational biology, including Critical Assessment of Structure Prediction (CASP) and Drug Design Data Resource (D3R) Grand Challenges.

Reactive molecular dynamics simulations show how ion-paring, cross-correlated ion motions and proton transfer contribute to the ionic conductivity in concentrated NaOH solutions at elevated temperatures.

The structural, electronic and thermal transport characteristics of novel carbon-based bilayer tetragonal graphene are systematically explored with a combination of first-principles calculations and machine-learning interatomic potential approaches.

Simulated research landscapes are used to create data mimicking experimental datasets. We populate the landscapes with different exploration strategies, and compare them to predict the benefit of using particular machine learning-guided strategies.

This study proposed a fragment-based graph convolutional neural network (F-GCN) that can predict atomic and inter-atomic properties and is suitable for few-shot learning.

Two high-precision water models are established based on the combination of a back-propagation neural network and genetic algorithm.

We demonstrate the power of our deep neural network to predict the X-ray absorption spectra of disordered systems in the presence of thermal energy.

We report here a global and full dimensional neural network potential energy surface for the F + CH₄ reaction and investigate the isotopic effect on the total reaction probabilities using the time-dependent wave packet method.

The present work reports a new full-dimensional potential energy surface (PES) of the HCl–H₂O dimer, and the first fully coupled 9D quantum calculations of the intra- and intermolecular vibrational states of the complex, utilizing this PES.

A new global potential energy surface (PES) for the ground state of the SH₂⁺(X⁴A′′) system is constructed using a permutation invariant polynomial neural network method.

Proton transfer along the hydrogen bonds of DNA can lead to the creation of short-lived, but biologically relevant point mutations that can further lead to gene mutation and, potentially, cancer.

The negatively activated region in CH₄ dissociation is attributed to a precursor-mediated mechanism involving surface defects.

Machine learning models were developed for an organic reaction in ionic liquids and validated on a selection of ionic liquids.

Data-augmented high-throughput QM approach to compute pK_a of transition metal hydride complexes with hDFT accuracy and low cost.

Combining piecewise switching functions with embedded atom neural networks to accelerate atomistic simulations with ab initio accuracy.

A random-forest machine learning classifier promotes an efficiency enhancement in the DFT re-optimization of microsolvation clusters by selecting promising minimum structures that were searched by an evolutionary algorithm on an analytical PES.

The FI-NN approach is capable of representing highly accurate diabatic PESs with particular and complicated symmetry problems.

In this study, we assess the application of a generative model to the NMDAR and provide source code for a variety of ligand- and structure-based assessment techniques used in standard drug discovery analyses to the deep learning-generated compounds.

An accurate full-dimensional PES for the OH + SO ↔ H + SO₂ reaction is developed by the permutation invariant polynomial-neural network approach.

A machine learning framework that generates material compositions exhibiting properties desired by the user.

We formulate a reaction prediction problem in terms of node-classification in a disconnected graph of source molecules and generalize a graph convolution neural network for disconnected graphs.

We present a first principles-quality neural-network potential energy surface describing interactions for a hydrogen atom with free-standing graphene.

We present a machine learning approach based on artificial neural networks for the prediction of ion pair solvation energies.

We develop a deep learning architecture to predict 3D transition state geometries. The new method generates structures more rapidly than existing QM methods.

Deep learning based methods have been widely applied to predict various kinds of molecular properties in the pharmaceutical industry with increasingly more success.

An iterative, combined experimental and computational approach towards predicting reaction rate constants in ionic liquids is presented.

Deep neural networks are a cost-effective machine-learning approach for solving the inverse problem of constructing electromagnetic fields that enable desired transitions in quantum systems.

Biomolecules have complex structures, and noncovalent interactions are crucial to determine their conformations and functionalities.

Machine-learning models enable important substructure detection and property prediction for drug–membrane interactions.

Machine learning algorithms can identify fluid and gel conformation states of individual lipid molecules.

Trained neural networks are used to extract the first partial coordination numbers from XANES spectra. In bimetallic nanoparticles, the four local structure descriptors provide rich information on structural motifs.

Based on the deep metric learning, a machine-guided representation is automatically optimized for any given dataset of molecular properties.

The developed deep neural network (DNN) potential can describe the structural properties of the Al₉₀Tb₁₀ liquid and the formation energies of Al–Tb crystals with the accuracy of ab initio calculations.

Graph neural networks with local and global attention mechanisms help to extract better features for materials property prediction.

Two manifold learning methods, isometric feature mapping and locally linear embedding, are applied to the analysis of quasi-classical trajectories for multi-channel reaction NH⁺ + H₂ → N + H₃⁺/NH₂⁺ + H.

The reaction and quenching processes of the K(4p²P) + H₂ reaction are studied based on new diabatic PESs.

Schematic of the developed neural network potential energy surface enabling a unified and transferable description of dynamics of H₂ dissociative adsorption on multiple copper surfaces.

We develop and test a framework for selecting appropriate chemical datasets to create molecular representations tailored for specific tasks.

Predicting phase stabilities of crystal polymorphs is central to computational materials science and chemistry.

The Solvent Similarity Index (SSI) is a quantitative parameter we introduce for the comparison of the solvation properties of any solvent or solvent mixture.

Using deep neural networks to model the polarizability and potential energy surfaces, we compute the Raman spectrum of liquid water at several temperatures with ab initio molecular dynamics accuracy.

The effects of intrinsic structural properties on the photoelectrochemical oxidation of phenolic pollutants at nanoporous TiO₂ are systemically studied.

Amorphous CaO–SiO₂–Al₂O₃ (CSA) slags are widely used in the glass, ceramic, cement and metallurgy industries.

Illustration of the calculation process to predict the relative stability of a protein upon point mutation from residue X to Y.

The kinetics methane activation (MgO⁺ + CH₄) was studied experimentally and computationally by running and analyzing reactive atomistic simulations.

Recently, low-dimensional mathematical representations have overshadowed other methods in drug discovery. This work reassesses eight 2D fingerprints on 23 molecular datasets and reveals that they can perform as well as mathematical representations in tasks involving only small molecules.

The reaction mechanisms of OH⁻ + D₂ → HOD + D⁻ were first revealed by theory, based on an accurate full-dimensional PES.

By ab initio molecular dynamics simulations, the newly developed SCAN meta-GGA functional is proved better than the widely used PBE-GGA functional in describing the equation of state of water.

The reinforcement learning method of double deep-Q learning is used to design moth-eye structure-based ultra-broadband perfect absorbers with a variety of transition metals, using transfer learning to share knowledge between different environments.

Based on unsupervised deep learning algorithms, an automatic data analysis method for single-molecule charge transport data is developed, which offers an opportunity to reveal more physical and chemical phenomena at the single-molecule level.

Thermal rate coefficients for the Cl + CH₄/CD₄ reactions were studied on a new full-dimensional accurate potential energy surface with the spin–orbit corrections considered in the entrance channel.