Artificial intelligence chemistry最新文献

Molecular similarity pervades much of our understanding and rationalization of chemistry. This has become particularly evident in the current data-intensive era of chemical research, with similarity measures serving as the backbone of many Machine Learning (ML) supervised and unsupervised procedures. Here, we present a discussion on the role of molecular similarity in drug design, chemical space exploration, chemical “art” generation, molecular representations, and many more. We also discuss more recent topics in molecular similarity, like the ability to efficiently compare large molecular libraries.

Large Language Models (LLMs), such as GPT-4, are precipitating a new "industrial revolution" by significantly enhancing productivity across various domains. These models encode an extensive corpus of scientific knowledge from vast textual datasets, functioning as near-universal generalists with the ability to engage in natural language communication and exhibit advanced reasoning capabilities. Notably, agents derived from LLMs can comprehend user intent and autonomously design, plan, and utilize tools to execute intricate tasks. These attributes are particularly advantageous for materials science research, an interdisciplinary field characterized by numerous complex and time-intensive activities. The integration of LLMs into materials science research holds the potential to fundamentally transform the research paradigm in this field.

Molecular conformations generation (MCG) aims to efficiently obtain reasonable and stable three-dimensional (3D) atomic coordinates of the atoms in the molecule from scratch, providing a structural foundation for molecular representation learning models and advanced downstream molecular design tasks such as molecular property prediction, molecular generation, and molecular docking. Existing MCG methods mostly rely on indirect distance-based strategies, which which can result in geometrically unrealistic conformations, or direct coordinate-based methods, which have larger search spaces and are prone to overfitting. Therefore, this study introduces Conf-GEM, a novel geometric information-assisted direct conformation generation model based on E-GeoGNN, a geometrically augmented 3D graph neural network with multiple scales. Pre-training and divide-and-conquer strategies, are integrated into the proposed model. Conf-GEM outperforms RDKit and nine deep-learning-based MCG models on the GEOM-QM9 and GEOM-Drugs datasets, achieving conformational coverage of 96.69% and 96.07%, respectively, without force field optimization. It also excels on the X-ray diffraction crystal structure dataset with up to 97.04% conformational coverage. In conclusion, Conf-GEM provides a novel solution for stabilizing 3D conformations generation. We provide an online prediction service (https://confgem.cmdrg.com) with a user-friendly interface for researchers.

The beginning and ripening of digital chemistry is analyzed focusing on the role of artificial intelligence (AI) in an expected leap in chemical sciences to bring this area to the next evolutionary level. The analytic description selects and highlights the top 20 AI-based technologies and 7 broader themes that are reshaping the field. It underscores the integration of digital tools such as machine learning, big data, digital twins, the Internet of Things (IoT), robotic platforms, smart control of chemical processes, virtual reality and blockchain, among many others, in enhancing research methods, educational approaches, and industrial practices in chemistry. The significance of this study lies in its focused overview of how these digital innovations foster a more efficient, sustainable, and innovative future in chemical sciences. This article not only illustrates the transformative impact of these technologies but also draws new pathways in chemistry, offering a broad appeal to researchers, educators, and industry professionals to embrace these advancements for addressing contemporary challenges in the field.

Artificial intelligence has brought crucial changes to the whole field of natural sciences. Myriads of machine learning algorithms have been developed to facilitate the work of experimental scientists. Molecular property prediction and drug synthesis planning become routine tasks. Moreover, inverse design of compounds with tunable properties as well as on-the-fly autonomous process optimization and chemical space exploration became possible in silico. Affordable robotic platforms exist able to perform thousands of experiments every day, analyzing the results and tuning the protocols. Despite this, most of these developments get trapped at the stage of code or overlooked, limiting their use by experimental scientists. Meanwhile, visibility and the number of user-friendly tools and technologies available to date is too low to compensate for this fact, rendering the development of novel therapeutic compounds inefficient. In this Review, we set the goal to bridge the gap between modern technologies and experimental scientists to improve drug development efficacy. Here we survey advanced and easy-to-use technologies able to help medical chemists at every stage of their research, including those integrated in technological processes during COVID-19 pandemic motivated by the need for fast yet precise solutions. Moreover, we review how these technologies are integrated by industry and clinics to streamline drug development and production. These technologies already transform the current paradigm of scientific thinking and revolutionize not only medicinal chemistry, but the whole field of natural sciences.

In this investigation, a quantitative structure-property relationship (QSPR) model coupled with a quantum neural network (QNN) was used to explore the corrosion inhibition efficiency (CIE) of quinoxaline compounds. Integrating quantum chemical properties (QCP) features reduced computational burden by strategically reducing the features from 11 to 4 while maintaining prediction accuracy. QNN models outperform traditional methods like artificial neural networks (ANN) and multilayer perceptron neural networks (MLPNN), with a coefficient of determination (R²) value of 0.987, coupled with diminished root mean square error (RMSE), mean absolute error (MAE), and mean absolute deviation (MAD) values of 0.97, 0.92, and 1.10, respectively. Predictions for six newly synthesized quinoxaline derivatives: quinoxaline-6-carboxylic acid (Q1), methyl quinoxaline-6-carboxylate (Q2), (2E,3E)-2,3-dihydrazono-1,2,3,4-tetrahydroquinoxaline (Q3), (2E,3E) 2,3-dihydrazono-6-methyl-1,2,3,4-tetrahydroquinoxaline (Q4), (E)-3-(4-methoxyethyl)-7-methylquinoxalin-2(1 H)-one (Q5), and 2-(4-methoxyphenyl)-7-methylthieno[3,2-b] quinoxaline (Q6), show remarkable CIE values of 95.12, 96.72, 91.02, 92.43, 89.58, and 93.63 %, respectively. This breakthrough technique simplifies testing and production procedures for new anti-corrosion materials.

Traditional drug discovery struggles to keep pace with the ever-evolving threat of infectious diseases. New viruses and antibiotic-resistant bacteria, all demand rapid solutions. Artificial Intelligence (AI) offers a promising path forward through accelerated drug repurposing. AI allows researchers to analyze massive datasets, revealing hidden connections between existing drugs, disease targets, and potential treatments. This approach boasts several advantages. First, repurposing existing drugs leverages established safety data and reduces development time and costs. Second, AI can broaden the search for effective therapies by identifying unexpected connections between drugs and potential new targets. Finally, AI can help mitigate limitations by predicting and minimizing side effects, optimizing drugs for repurposing, and navigating intellectual property hurdles. The article explores specific AI strategies like virtual screening, target identification, structure base drug design and natural language processing. Real-world examples highlight the potential of AI-driven drug repurposing in discovering new treatments for infectious diseases.

Drug Research and Development (R&D) is a complex and difficult process, and current drug R&D faces the challenges of long time span, high investment, and high failure rate. Machine learning, with its powerful learning ability to characterize big data and complex networks, is increasingly effective to improve the efficiency and success rate of drug R&D. Here we review some recent examples of the application of machine learning methods in six areas: disease gene prediction, virtual screening, drug molecule generation, molecular attribute prediction, and prediction of drug combination synergism. We also discuss the advantages of integrative learning in multi-attribute prediction. Integrative models based on base learners constructed from data of different dimensions on the one hand fully utilize the information contained in these data, and on the other hand improve the average prediction performance. Finally, we envision a new paradigm for drug discovery and development: a large language model acts as a central hub to organize public resources into a knowledge base, validating the knowledge with computational software and smaller predictive models, as well as high-throughput automated screening platforms based on organoidal technologies, to speed up development and reduce the differences in efficacy between disease models and humans to improve the success rate of a drug.

Methanol-to-olefins (MTO), as an alternative pathway for the synthesis of light olefins (ethylene and propylene), has gained extensive attention. Accurate prediction of light olefins yields can effectively facilitate process monitoring and optimization, as they are significant economic indexes and stable operation indicators of the industrial MTO process. However, the nonlinearity and dynamic interactions among process variables pose challenges for the prediction using traditional statistical methods. Additionally, physical-based methods relying on first-principle theory are always limited by an insufficient understanding of reaction mechanisms. In contrast, data-driven methods offer a viable solution for the prediction based solely on process data without requiring extensive process knowledge. Therefore, in this work, a data-driven approach that integrates spatial and temporal self-attention modules is proposed to capture complex interactions. Furthermore, Bayesian optimization is employed to determine the optimum hyperparameters and enhance the accuracy of the model. Studies on an actual MTO process demonstrate the superior prediction performance of the proposed model compared to baseline models. Specifically, 24 process variables are selected as the high-dimensional inputs, and yields of ethylene and propylene, as the low-dimensional outputs, are successfully predicted at various prediction horizons ranging from 2 to 8 h.

Activity of water and electrolytes in aqueous solutions is of utmost importance for multiple industrial applications. However, experimental determination of such values is time-consuming, while calculation of activity coefficients using numerical methods is challenging. By training neural networks models on literature data, one could predict activity of water and electrolytes easily, without requiring any experiment. In this paper, multiple descriptors (or features) are compared to predict activity coefficients of electrolytes and activity of water in electrolyte solutions. A neural network based on the Levenberg-Marquardt algorithm (LM-NN) showed high accuracy to calculate values, despite the small size of the training datasets. Both activity coefficients of electrolytes and activity of water in electrolyte solutions can be predicted accurately even on unseen data, using simple descriptors such as electrolyte concentration, ion sizes and charges. However, some discrepancies were observed due to the lack of representativeness of the training dataset. This could be solved by selecting training data sets that are similar (e.g. same group of the periodic table) to the unknown values, or by including available experimental data for the salt considered. The ability of the LM-NN to solve non-linear least square curve fitting problems makes it a good candidate to fit experimental activity coefficient data, with the advantage of simplicity as compared to e-NRTL or UNIQUAC methods. This method paves the way for accurate and quick determination of thermodynamic data for electrolyte solutions (and beyond) using machine learning, without necessitating large training datasets.