Graph neural networks for materials science and chemistry

Communications Materials, Journal Year: 2022, Volume and Issue: 3(1)

Published: Nov. 26, 2022

Abstract Machine learning plays an increasingly important role in many areas of chemistry and materials science, being used to predict properties, accelerate simulations, design new structures, synthesis routes materials. Graph neural networks (GNNs) are one the fastest growing classes machine models. They particular relevance for as they directly work on a graph or structural representation molecules therefore have full access all relevant information required characterize In this Review, we provide overview basic principles GNNs, widely datasets, state-of-the-art architectures, followed by discussion wide range recent applications GNNs concluding with road-map further development application GNNs.

Language: Английский

---

InteractionGraphNet: A Novel and Efficient Deep Graph Representation Learning Framework for Accurate Protein–Ligand Interaction Predictions

Journal of Medicinal Chemistry, Journal Year: 2021, Volume and Issue: 64(24), P. 18209 - 18232

Published: Dec. 8, 2021

Accurate quantification of protein–ligand interactions remains a key challenge to structure-based drug design. However, traditional machine learning (ML)-based methods based on handcrafted descriptors, one-dimensional protein sequences, and/or two-dimensional graph representations limit their capability learn the generalized molecular in 3D space. Here, we proposed novel deep representation framework named InteractionGraphNet (IGN) from structures complexes. In IGN, two independent convolution modules were stacked sequentially intramolecular and intermolecular interactions, learned can be efficiently used for subsequent tasks. Extensive binding affinity prediction, large-scale virtual screening, pose prediction experiments demonstrated that IGN achieved better or competitive performance against other state-of-the-art ML-based baselines docking programs. More importantly, such was proven successful features instead just memorizing certain biased patterns data.

Language: Английский

---

A multi-factor driven spatiotemporal wind power prediction model based on ensemble deep graph attention reinforcement learning networks

Energy, Journal Year: 2022, Volume and Issue: 263, P. 126034 - 126034

Published: Nov. 11, 2022

Language: Английский

---

An improved GNN using dynamic graph embedding mechanism: A novel end-to-end framework for rolling bearing fault diagnosis under variable working conditions

Mechanical Systems and Signal Processing, Journal Year: 2023, Volume and Issue: 200, P. 110534 - 110534

Published: June 21, 2023

Language: Английский

---

Deep learning predicts boiling heat transfer

Scientific Reports, Journal Year: 2021, Volume and Issue: 11(1)

Published: March 10, 2021

Boiling is arguably Nature's most effective thermal management mechanism that cools submersed matter through bubble-induced advective transport. Central to the boiling process development of bubbles. Connecting physics with bubble dynamics an important, yet daunting challenge because intrinsically complex and high dimensional dynamics. Here, we introduce a data-driven learning framework correlates high-quality imaging on dynamic bubbles associated curves. The leverages cutting-edge deep models including convolutional neural networks object detection algorithms automatically extract both hierarchical physics-based features. By training these features, our model learns physical laws statistically describe manner in which nucleate, coalesce, depart under conditions, enabling situ curve prediction mean error 6%. Our offers automated, learning-based, alternative conventional heat transfer metrology.

Language: Английский

---

RELATION: A Deep Generative Model for Structure-Based De Novo Drug Design

Journal of Medicinal Chemistry, Journal Year: 2022, Volume and Issue: 65(13), P. 9478 - 9492

Published: June 17, 2022

Deep learning (DL)-based de novo molecular design has recently gained considerable traction. Many DL-based generative models have been successfully developed to novel molecules, but most of them are ligand-centric and the role 3D geometries target binding pockets in generation not well-exploited. Here, we proposed a new 3D-based model called RELATION. In RELATION model, BiTL algorithm was specifically designed extract transfer desired geometric features protein-ligand complexes latent space for generation. The pharmacophore conditioning docking-based Bayesian sampling were applied efficiently navigate vast chemical molecules with properties features. As proof concept, used inhibitors two targets, AKT1 CDK2. calculation results demonstrated that could generate favorable affinity

Language: Английский

---

A new ensemble deep graph reinforcement learning network for spatio-temporal traffic volume forecasting in a freeway network

Digital Signal Processing, Journal Year: 2022, Volume and Issue: 123, P. 103419 - 103419

Published: Jan. 29, 2022

Language: Английский

---

Review of machine learning and deep learning models for toxicity prediction

Experimental Biology and Medicine, Journal Year: 2023, Volume and Issue: unknown

Published: Dec. 6, 2023

The ever-increasing number of chemicals has raised public concerns due to their adverse effects on human health and the environment. To protect environment, it is critical assess toxicity these chemicals. Traditional in vitro vivo assays are complicated, costly, time-consuming may face ethical issues. These constraints raise need for alternative methods assessing Recently, advancement machine learning algorithms increase computational power, many prediction models have been developed using various deep such as support vector machine, random forest, k-nearest neighbors, ensemble learning, neural network. This review summarizes learning- learning-based recent years. Support forest most popular algorithms, hepatotoxicity, cardiotoxicity, carcinogenicity frequently modeled endpoints predictive toxicology. It known that datasets impact model performance. quality used development vital performance models. different assignments same among type observed, indicating benchmarking needed developing reliable algorithms. provides insights into current toxicology, which expected promote application future.

Language: Английский

---

Discovery of Pyrazine-Carboxamide-Diphenyl-Ethers as Novel Succinate Dehydrogenase Inhibitors via Fragment Recombination

Journal of Agricultural and Food Chemistry, Journal Year: 2020, Volume and Issue: 68(47), P. 14001 - 14008

Published: Nov. 13, 2020

The discovery of novel succinate dehydrogenase inhibitors (SDHIs) has attracted great attention worldwide. Herein, a fragment recombination strategy was proposed to design new SDHIs by understanding the ligand–receptor interaction mechanism SDHIs. Three fragments, pyrazine from pyraziflumid, diphenyl-ether flubeneteram, and prolonged amide linker pydiflumetofen fluopyram, were identified recombined produce pyrazine-carboxamide-diphenyl-ether scaffold as SDHI. After substituent optimization, compound 6y successfully with good inhibitory activity against porcine SDH, which about 2-fold more potent than pyraziflumid. Furthermore, exhibited 95% 80% rates soybean gray mold wheat powdery mildew at dosage 100 mg/L in vivo assay, respectively. results present work showed that could be used starting point for

Language: Английский

---

Cloud 3D-QSAR: a web tool for the development of quantitative structure–activity relationship models in drug discovery

Briefings in Bioinformatics, Journal Year: 2020, Volume and Issue: 22(4)

Published: Sept. 22, 2020

Abstract Effective drug discovery contributes to the treatment of numerous diseases but is limited by high costs and long cycles. The Quantitative Structure–Activity Relationship (QSAR) method was introduced evaluate activity a large number compounds virtually, reducing time labor required for chemical synthesis experimental determination. Hence, this increases efficiency discovery. To meet needs researchers utilize technology, QSAR-related web servers, such as Web-4D-QSAR DPubChem, have been developed in recent years. However, none servers mentioned above can perform complete QSAR modeling supply prediction functions. We introduce Cloud 3D-QSAR integrating functions molecular structure generation, alignment, interaction field (MIF) computing results analysis provide one-stop solution. rigidly validated server, correlation R2 = 0.934 834 test molecules. sensitivity, specificity accuracy were 86.9%, 94.5% 91.5%, respectively, with AUC 0.981, AUCPR 0.971. server may facilitate development good models Our free now available at http://chemyang.ccnu.edu.cn/ccb/server/cloud3dQSAR/ http://agroda.gzu.edu.cn:9999/ccb/server/cloud3dQSAR/.

Language: Английский

---