Artificial intelligence chemistry最新文献

In this work, the previously introduced NeuralSchrödinger PINN is extended towards the use of analytical gradient expressions of the loss function. It is shown that the analytical gradients derived in this work increase the convergence properties for both the BFGS and ADAM optimizers compared to the previously employed numerical gradient implementation. In addition, the use of parallelised GPU computations via CUDA greatly increased the computational performance over the previous implementation using single-core CPU computations. As a consequence, an extension of the NeuralSchrödinger PINN towards two-dimensional quantum systems became feasible as also demonstrated in this work.

Synchrotron radiation technology provides high-resolution and high-sensitivity information for many fields such as material science, life science, and energy research. Synchrotron radiation data-driven methods have significantly accelerated the development of materials discovery and analysis. However, synchrotron radiation data is complex and large, requiring artificial intelligence for analysis. Artificial intelligence can efficiently process complex high-dimensional data, automate the analysis process, discover hidden patterns and associations, and build predictive models. This review provides an overview of the application and development of combining synchrotron radiation data-driven methods with artificial intelligence in the field of materials discovery. The application of the method in science is still limited by the problems of large and complex synchrotron radiation data, valuable experimental machine time, and uninterpretable artificial intelligence models. To address these problems, this review correspondingly proposes solutions for synchrotron radiation artificial intelligence data banks, standardized experiment records systems, and interpretable artificial intelligence predictive models.

The interaction between water and graphene is important for understanding the thermodynamic and kinetic properties of water on hydrophobic surfaces. In this study, we constructed a high-dimensional potential energy surface (PES) for the water-graphene system using the many-body expansion scheme and neural network fitting. By analyzing the landscape of the PES, we found that the water molecule exhibits a weak physisorption behavior with a binding energy of about − 1000 cm⁻¹ and a very low diffusion barrier. Furthermore, extensive molecular dynamics were performed to investigate the adsorption and diffusion dynamics of a single water on a graphene surface at temperatures ranging from 50 to 300 K. Potential-of-mean-forces were computed from the trajectories, providing a comprehensive and accurate description of the water-graphene interaction kinetics.

Targeting riboswitches, regulatory elements responsible for the expression of essential genes, is taking central stage in the new era of antibacterial medications discovery due to the emergence of antibiotic resistance. The S-Adenosyl methionine-I (SAM-I) riboswitch works through transcription termination in a negative feedback manner modulated by the natural ligand SAM. SAM-I riboswitch is specific to bacteria and found mainly in gram-positive bacteria such as Bacillus anthracis. Analyzing the interactions of the co-crystallized structure of SAM-I riboswitch aptamer with its native ligand SAM clarified the needed chemical structural features to achieve binding. Acknowledging those features, structure-based and ligand-based pharmacophore models were built for filtration use in screening the OTAVA Chemical library and the Pubchem database. For further filtration enhancement, the physicochemical properties of SAM were used as a second filtration criterion. Compounds obtained as output from previous steps were energy minimized, and the lowest energy conformer structures were docked to SAM-I using MOE, v.2019.01. S-score and ligand interactions were used to assess the best hits. This yielded eight promising compounds to which molecular dynamics (MD) simulations with SAM-I aptamer were applied using GROMACS 2020.3 package affirming stable binding interactions and binding energetics similar to SAM. Moreover, pharmacokinetic and drug-like properties of those eight hits were assessed using SWISS-ADME. According to the combined computational methods and PK/Tox assessment, compound 20 was the most promising and thus can be considered a lead for future evaluation and optimization as a candidate new antibacterial agent targeting a new biomolecule eliciting a new mechanism of action.

Fluorine-19 (¹⁹F) is a nucleus of great importance in the field of Nuclear Magnetic Resonance (NMR) spectroscopy due to its high receptivity and wide chemical shift dispersion. ¹⁹F NMR plays crucial roles in both organic synthesis and biomedicine. Herein, a machine learning-based comprehensive ¹⁹F NMR chemical shift prediction model was established based on the experimental ¹⁹F NMR dataset from the book by Dolbier and the open NMR database nmrshiftdb2. Fluorine radical SMILES (Fr-SMILES) that reflected the fluorine chemical equivalence, was designed as the representation of fluorine in the molecule. Model trained with the graph convolution network (GCN) algorithm gave a low mean absolute error (MAE) of 3.636 ppm on the testing set. This model exhibits broad applicability and can effectively predict ¹⁹F NMR shifts for a wide range of organic fluorine molecules. We believe that the current work will provide a powerful tool for not only predicting ¹⁹F NMR shifts but also aiding in the analysis and identification of these shifts in diverse organic fluorine compounds. An online prediction platform was constructed based on the current model, which can be found at https://fluobase.cstspace.cn/fnmr.

Single-atom catalysts (SACs), featuring isolated metal atoms embedded in graphitic carbon materials, have attracted considerable research interest due to their cost-effectiveness, high catalytic activity, and customizable functionality across various catalytic reactions. Among SACs, the Fe-N₄-C class has garnered significant attention. Tailoring the properties of Fe-N₄ sites through localized chemical modifications stands as a key strategy for catalyst engineering. Recent experimental and computational investigations have underscored the distinct influence of axial ligands on Fe in modulating the oxygen reduction reaction (ORR) activity. However, the precise quantitative structure-property relationship between ligands and the catalytic properties of the Fe center remains elusive. In this study, we combined the density functional theory (DFT) simulations and machine learning (ML) models to unravel the relationship between the ligand properties and the oxo binding energy. This energy pertains to the binding of an oxygen atom to the Fe center, a fundamental step in ORR. Through the design of 33 ligands and 5 molecular complexes that accommodate the Fe-N₄ moiety, we screened a total of 278 oxo binding energies across an array of ligands and host complexes. Harnessing the power of ML models, we achieved an accurate prediction of these oxo binding energies using features collected from DFT simulations. Notably, the predominant features contributing to the oxo binding energy prediction primarily derived from complexes with attached ligands, rather than isolated ligand properties. We formulated an approach that leverages these critical features and identified the isolated ligand properties capable of effectively predicting these features. This methodology can potentially be applied to investigate other ORR intermediates and a comprehensive understanding of the ligand effect for the ORR activity in SACs can be achieved.

Despite uranium dioxide (UO₂) being a widely used nuclear fuel, fuel performance models rely extensively on empirical correlations of material behavior, leveraging the historical operating experience of UO₂. Mechanistic models that consider an atomistic understanding of the processes governing fuel performance (such as fission gas release and creep) will enable a better description of fuel behavior under non-prototypical conditions such as in new reactor concepts or for modified UO₂ fuel compositions. To this end, molecular dynamics simulation is a powerful tool for rapidly predicting physical properties of proposed fuel candidates. However, the reliability of these simulations depends largely on the accuracy of the atomic forces. Traditionally, these forces are computed using either a classical force field (FF) or density functional theory (DFT). While DFT is relatively accurate, the computational cost is burdensome, especially for f-electron elements, such as actinides. By contrast, classical FFs are computationally efficient but are less accurate. For these reasons, we report a new accurate machine learning interatomic potential (MLIP) for UO₂ that provides high-fidelity reproduction of DFT forces at a similar low cost to classical FFs. We employ an active learning approach that autonomously augments the DFT training data set to iteratively refine the MLIP. To further improve the quality of our predictions, we utilize transfer learning to retrain our MLIP to higher-accuracy DFT+U data. We validate our MLIPs by comparing predicted physical properties (e.g., thermal expansion and elastic properties) with those from existing classical FFs and DFT/DFT+U calculations, as well as with experimental data when available.

The orexin 1 receptor (OX1R) is a G-protein coupled receptor that regulates a variety of physiological processes through interactions with the neuropeptides orexin A and B. Selective OX1R antagonists exhibit therapeutic effects in preclinical models of several behavioral disorders, including drug seeking and overeating. However, currently there are no selective OX1R antagonists approved for clinical use, fueling demand for novel compounds that act at this target. In this study, we meticulously curated a dataset comprising over 1300 OX1R ligands using a stringent filter and criteria cascade. Subsequently, we developed highly predictive quantitative structure-activity relationship (QSAR) models employing the optimized hyper-parameters for the random forest machine learning algorithm and twelve 2D molecular descriptors selected by recursive feature elimination with a 5-fold cross-validation process. The predictive capacity of the QSAR model was further assessed using an external test set and enrichment study, confirming its high predictivity. The practical applicability of our final QSAR model was demonstrated through virtual screening of the DrugBank database. This revealed two FDA-approved drugs (isavuconazole and cabozantinib) as potential OX1R ligands, confirmed by radiolabeled OX1R binding assays. To our best knowledge, this study represents the first report of highly predictive QSAR models on a large comprehensive dataset of diverse OX1R ligands, which should prove useful for the discovery and design of new compounds targeting this receptor.

Artificial intelligence (AI) is revolutionizing the current process of drug design and development, addressing the challenges encountered in its various stages. By utilizing AI, the efficiency of the process is significantly improved through enhanced precision, reduced time and cost, high-performance algorithms and AI-enabled computer-aided drug design (CADD). Effective drug screening techniques are crucial for identifying potential hit compounds from large volumes of data in compound repositories. The inclusion of AI in drug discovery, including the screening of hit compounds and lead molecules, has proven to be more effective than traditional in vitro screening assays. This article reviews the advancements in drug screening methods achieved through AI-enhanced applications, machine learning (ML), and deep learning (DL) algorithms. It specifically focuses on AI applications in the drug discovery phase, exploring screening strategies and lead optimization techniques such as Quantitative structure-activity relationship (QSAR) modeling, pharmacophore modeling, de novo drug designing, and high-throughput virtual screening. Valuable insights into different aspects of the drug screening process are discussed, highlighting the role of AI-based tools, pipelines, and case studies in simplifying the complexities associated with drug discovery.

Developing new pharmaceutical compounds is a lengthy, costly, and intensive process. In recent years, the development of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) models has drawn considerable interest in drug discovery. In this review, we discuss recent advances in the field and show how these methods can be leveraged to assist each stage of the drug discovery process. After discussing recent technical progress in the encoding of chemical information via fingerprinting and the emergence of graph-based and generative models, we examine all types of interactions, including drug-target interactions, protein-protein interactions, protein-peptide interactions, and nucleic acid-based interactions. Furthermore, we discuss recent advances enabled by DL models for the prediction of ADMET (Absorption, Distribution, Metabolism, Elimination, Toxicity) properties and of solubility. We also review applications that have emerged in the past two years with the development of models, for instance, on SARS-CoV-2 inhibitors and highlight outstanding challenges.