Could graph neural networks learn better molecular representation for drug discovery? A comparison study of descriptor-based and graph-based models

Journal of Cheminformatics, Journal Year: 2021, Volume and Issue: 13(1)

Published: Feb. 17, 2021

Abstract Graph neural networks (GNN) has been considered as an attractive modelling method for molecular property prediction, and numerous studies have shown that GNN could yield more promising results than traditional descriptor-based methods. In this study, based on 11 public datasets covering various endpoints, the predictive capacity computational efficiency of prediction models developed by eight machine learning (ML) algorithms, including four (SVM, XGBoost, RF DNN) graph-based (GCN, GAT, MPNN Attentive FP), were extensively tested compared. The demonstrate average outperform in terms accuracy efficiency. SVM generally achieves best predictions regression tasks. Both XGBoost can achieve reliable classification tasks, some models, such FP GCN, outstanding performance a fraction larger or multi-task datasets. cost, are two most efficient algorithms only need few seconds to train model even large dataset. interpretations SHAP effectively explore established domain knowledge models. Finally, we explored use these virtual screening (VS) towards HIV demonstrated different ML offer diverse VS profiles. All all, believe off-the-shelf still be directly employed accurately predict chemical endpoints with excellent computability interpretability.

Language: Английский

---

Predicting student satisfaction of emergency remote learning in higher education during COVID-19 using machine learning techniques

PLoS ONE, Journal Year: 2021, Volume and Issue: 16(4), P. e0249423 - e0249423

Published: April 2, 2021

Despite the wide adoption of emergency remote learning (ERL) in higher education during COVID-19 pandemic, there is insufficient understanding influencing factors predicting student satisfaction for this novel environment crisis. The present study investigated important predictors determining undergraduate students (N = 425) from multiple departments using ERL at a self-funded university Hong Kong while Moodle and Microsoft Team are key tools. By comparing predictive accuracy between regression machine models before after use random forest recursive feature elimination, all regression, showed improved most accurate model was elastic net with 65.2% explained variance. results show only neutral (4.11 on 7-point Likert scale) regarding overall score ERL. Even majority competent technology have no obvious issue accessing devices or Wi-Fi, face-to-face more preferable compared to found be predictor. Besides, level efforts made by instructors, agreement appropriateness adjusted assessment methods, perception online being well delivered shown highly scores. suggest that need reviewing quality quantity modified accommodated structured class delivery suitable amount interactive according culture program nature.

Language: Английский

---

Development of machine learning model for diagnostic disease prediction based on laboratory tests

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

Published: April 7, 2021

Abstract The use of deep learning and machine (ML) in medical science is increasing, particularly the visual, audio, language data fields. We aimed to build a new optimized ensemble model by blending DNN (deep neural network) with two ML models for disease prediction using laboratory test results. 86 attributes (laboratory tests) were selected from datasets based on value counts, clinical importance-related features, missing values. collected sample 5145 cases, including 326,686 investigated total 39 specific diseases International Classification Diseases, 10th revision (ICD-10) codes. These used construct light gradient boosting (LightGBM) extreme (XGBoost) TensorFlow. achieved an F1-score 81% accuracy 92% five most common diseases. showed differences predictive power classification patterns. confusion matrix analyzed feature importance SHAP method. Our high efficiency through This study will be useful diagnosis

Language: Английский

---

Assessing bikeability with street view imagery and computer vision

Transportation Research Part C Emerging Technologies, Journal Year: 2021, Volume and Issue: 132, P. 103371 - 103371

Published: Sept. 20, 2021

Language: Английский

---

Artificial Intelligence in Drug Toxicity Prediction: Recent Advances, Challenges, and Future Perspectives

Journal of Chemical Information and Modeling, Journal Year: 2023, Volume and Issue: 63(9), P. 2628 - 2643

Published: April 26, 2023

Toxicity prediction is a critical step in the drug discovery process that helps identify and prioritize compounds with greatest potential for safe effective use humans, while also reducing risk of costly late-stage failures. It estimated over 30% candidates are discarded owing to toxicity. Recently, artificial intelligence (AI) has been used improve toxicity as it provides more accurate efficient methods identifying potentially toxic effects new before they tested human clinical trials, thus saving time money. In this review, we present an overview recent advances AI-based prediction, including various machine learning algorithms deep architectures, six major properties Tox21 assay end points. Additionally, provide list public data sources useful tools research community highlight challenges must be addressed enhance model performance. Finally, discuss future perspectives prediction. This review can aid researchers understanding pave way discovery.

Language: Английский

---

A systematic review and meta-analysis of artificial neural network, machine learning, deep learning, and ensemble learning approaches in field of geotechnical engineering

Neural Computing and Applications, Journal Year: 2024, Volume and Issue: 36(21), P. 12655 - 12699

Published: May 13, 2024

Abstract Artificial neural networks (ANN), machine learning (ML), deep (DL), and ensemble (EL) are four outstanding approaches that enable algorithms to extract information from data make predictions or decisions autonomously without the need for direct instructions. ANN, ML, DL, EL models have found extensive application in predicting geotechnical geoenvironmental parameters. This research aims provide a comprehensive assessment of applications addressing forecasting within field related engineering, including soil mechanics, foundation rock environmental geotechnics, transportation geotechnics. Previous studies not collectively examined all algorithms—ANN, EL—and explored their advantages disadvantages engineering. categorize address this gap existing literature systematically. An dataset relevant was gathered Web Science subjected an analysis based on approach, primary focus objectives, year publication, geographical distribution, results. Additionally, study included co-occurrence keyword covered techniques, systematic reviews, review articles data, sourced Scopus database through Elsevier Journal, were then visualized using VOS Viewer further examination. The results demonstrated ANN is widely utilized despite proven potential methods engineering due real-world laboratory civil engineers often encounter. However, when it comes behavior scenarios, techniques outperform three other methods. discussed here assist understanding benefits geo area. enables practitioners select most suitable creating certainty resilient ecosystem.

Language: Английский

---

Risk-driven composition decoupling analysis for urban flooding prediction in high-density urban areas using Bayesian-Optimized LightGBM

Journal of Cleaner Production, Journal Year: 2024, Volume and Issue: 457, P. 142286 - 142286

Published: April 20, 2024

Language: Английский

---

Molecular property prediction: recent trends in the era of artificial intelligence

Drug Discovery Today Technologies, Journal Year: 2019, Volume and Issue: 32-33, P. 29 - 36

Published: Dec. 1, 2019

Language: Английский

---

GGL-Tox: Geometric Graph Learning for Toxicity Prediction

Journal of Chemical Information and Modeling, Journal Year: 2021, Volume and Issue: 61(4), P. 1691 - 1700

Published: March 15, 2021

Toxicity analysis is a major challenge in drug design and discovery. Recently significant progress has been made through machine learning due to its accuracy, efficiency, lower cost. US Toxicology the 21st Century (Tox21) screened large library of compounds, including approximately 12 000 environmental chemicals drugs, for different mechanisms responsible eliciting toxic effects. The Tox21 Data Challenge offered platform evaluate computational methods toxicity predictions. Inspired by success multiscale weighted colored graph (MWCG) theory protein-ligand binding affinity predictions, we consider MWCG analysis. In present work, develop geometric (GGL-Tox) model integrating features gradient boosting decision tree (GBDT) algorithm. benchmark tests are employed demonstrate utility usefulness proposed GGL-Tox model. An extensive comparison with other state-of-the-art models indicates that an accurate efficient prediction.

Language: Английский

---

Machine learning and artificial intelligence in physiologically based pharmacokinetic modeling

Toxicological Sciences, Journal Year: 2022, Volume and Issue: 191(1), P. 1 - 14

Published: Sept. 22, 2022

Physiologically based pharmacokinetic (PBPK) models are useful tools in drug development and risk assessment of environmental chemicals. PBPK model requires the collection species-specific physiological, chemical-specific absorption, distribution, metabolism, excretion (ADME) parameters, which can be a time-consuming expensive process. This raises need to create computational capable predicting input parameter values for models, especially new compounds. In this review, we summarize an emerging paradigm integrating modeling with machine learning (ML) or artificial intelligence (AI)-based methods. includes 3 steps (1) obtain time-concentration PK data and/or ADME parameters from publicly available databases, (2) develop ML/AI-based approaches predict (3) incorporate ML/AI into summary statistics (eg, area under curve maximum plasma concentration). We also discuss neural network architecture "neural ordinary differential equation (Neural-ODE)" that is providing better predictive capabilities than other ML methods when used directly time-series profiles. order support applications development, several challenges should addressed as more become available, it important expand training set by including structural diversity compounds improve prediction accuracy models; due black box nature many lack sufficient interpretability limitation; Neural-ODE has great potential generate profiles limited information, but its application remains explored. Despite existing challenges, will continue facilitate efficient robust large number

Language: Английский

---