Pub Date : 2026-01-23DOI: 10.1016/j.aichem.2026.100107
Jakes Udabe
Artificial intelligence (AI) is increasingly steering the discovery of functional molecules and materials, but its progress with generative modeling is held back by the messy, mixed-up nature of the experimental data and a scarcity of high-quality ground truth. This review synthesizes recent advances in data curation, representation, and generative modeling for molecular and materials discovery, and proposes a practical four-stage workflow that integrates structured data capture, intelligent featurization, generative design, and closed-loop experimental validation. Core algorithmic families (supervised, semi-supervised, unsupervised, reinforcement learning) and specialized generative architectures (VAEs, GANs, diffusion models, graph-based models) are surveyed, and discuss how each maps to real-world discovery tasks. The enabling infrastructure (e.g.as electronic lab notebooks (ELNs), knowledge graphs, autonomous laboratories) is likewise analyzed and highlight best practices for reproducibility, uncertainty quantification, and ethical safeguards. Finally, a prioritized checklist was provided for researchers and laboratories to adopt AI-compatible infrastructure and describe open challenges (data standards, causal inference, accessibility) to guide future work.
{"title":"A scientist's guide to AI-driven molecular discovery","authors":"Jakes Udabe","doi":"10.1016/j.aichem.2026.100107","DOIUrl":"10.1016/j.aichem.2026.100107","url":null,"abstract":"<div><div>Artificial intelligence (AI) is increasingly steering the discovery of functional molecules and materials, but its progress with generative modeling is held back by the messy, mixed-up nature of the experimental data and a scarcity of high-quality ground truth. This review synthesizes recent advances in data curation, representation, and generative modeling for molecular and materials discovery, and proposes a practical four-stage workflow that integrates structured data capture, intelligent featurization, generative design, and closed-loop experimental validation. Core algorithmic families (supervised, semi-supervised, unsupervised, reinforcement learning) and specialized generative architectures (VAEs, GANs, diffusion models, graph-based models) are surveyed, and discuss how each maps to real-world discovery tasks. The enabling infrastructure (e.g.as electronic lab notebooks (ELNs), knowledge graphs, autonomous laboratories) is likewise analyzed and highlight best practices for reproducibility, uncertainty quantification, and ethical safeguards. Finally, a prioritized checklist was provided for researchers and laboratories to adopt AI-compatible infrastructure and describe open challenges (data standards, causal inference, accessibility) to guide future work.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"4 1","pages":"Article 100107"},"PeriodicalIF":0.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077851","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-23DOI: 10.1016/j.aichem.2026.100108
Rahul D. Jawarkar , Prashant K. Deshmukh , Bhavesh Mandwale , Long Chaiou Ming
Generative AI and deep learning improve molecular simulations and drug development. Traditional computational methods like MD, MC, and QM/MM have been crucial in investigating biomolecular interactions and thermodynamics. However, processing power and speed restrict their scalability. This article provides a comprehensive review and comparative analysis of how advanced neural network architectures and generative AI models address these computational limitations. This review analyses how advanced neural network architectures and generative AI models satisfy these restrictions. Neural network potentials trained on high-quality quantum datasets achieve ab initio precision at low processing cost. We tested convolutional (CNNs), recurrent (RNNs), graph neural networks (GNNs), and transformers to evaluate how well they could describe molecular changes over time and predict structural changes. Researchers have investigated generative frameworks including variational autoencoders (VAEs), generative adversarial networks (GANs), and diffusion models to develop medications with superior binding affinity and pharmacokinetic characteristics. The findings reveal that AI-driven modelling and physics-based simulations create a closed-loop system where MD or QM/MM simulations enhance AI-generated molecules repeatedly. This feedback loop speeds up hit-to-lead optimisation, increases ADMET prediction, and enhances protein folding and shape information. This paradigm shift from descriptive to predictive and generative frameworks using AI and molecular modelling improves computational drug discovery's scalability, interpretability, and creativity. AI is used as a computational tool and a collaborator to speed up molecular discovery. Overall, this manuscript serves as a critical review summarizing state-of-the-art progress, challenges, and future prospects at the interface of AI and molecular simulation research.
{"title":"From potential to practice: The prospective and pitfalls of generative AI and deep learning in molecular simulations","authors":"Rahul D. Jawarkar , Prashant K. Deshmukh , Bhavesh Mandwale , Long Chaiou Ming","doi":"10.1016/j.aichem.2026.100108","DOIUrl":"10.1016/j.aichem.2026.100108","url":null,"abstract":"<div><div>Generative AI and deep learning improve molecular simulations and drug development. Traditional computational methods like MD, MC, and QM/MM have been crucial in investigating biomolecular interactions and thermodynamics. However, processing power and speed restrict their scalability. This article provides a comprehensive review and comparative analysis of how advanced neural network architectures and generative AI models address these computational limitations. This review analyses how advanced neural network architectures and generative AI models satisfy these restrictions. Neural network potentials trained on high-quality quantum datasets achieve ab initio precision at low processing cost. We tested convolutional (CNNs), recurrent (RNNs), graph neural networks (GNNs), and transformers to evaluate how well they could describe molecular changes over time and predict structural changes. Researchers have investigated generative frameworks including variational autoencoders (VAEs), generative adversarial networks (GANs), and diffusion models to develop medications with superior binding affinity and pharmacokinetic characteristics. The findings reveal that AI-driven modelling and physics-based simulations create a closed-loop system where MD or QM/MM simulations enhance AI-generated molecules repeatedly. This feedback loop speeds up hit-to-lead optimisation, increases ADMET prediction, and enhances protein folding and shape information. This paradigm shift from descriptive to predictive and generative frameworks using AI and molecular modelling improves computational drug discovery's scalability, interpretability, and creativity. AI is used as a computational tool and a collaborator to speed up molecular discovery. Overall, this manuscript serves as a critical review summarizing state-of-the-art progress, challenges, and future prospects at the interface of AI and molecular simulation research.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"4 1","pages":"Article 100108"},"PeriodicalIF":0.0,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146077763","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-02DOI: 10.1016/j.aichem.2025.100106
Eslam G. Al-Sakkari , Ahmed Ragab , Marzouk Benali , Olumoye Ajao , Daria C. Boffito , Hanane Dagdougui
Carbon capture, utilization and storage (CCUS), along with lignocellulosic biomass valorization (e.g., lignin, cellulose), are promising decarbonization strategies for hard-to-abate industries. Green solvents, such as deep eutectic solvents and ionic liquids, enable efficient CO₂ capture and selective lignin extraction, enhancing lignin depolymerization into high-value products. However, current molecular design tools are slow and computationally expensive, limiting green material innovation. This study introduces a novel data-driven framework for green material discovery using generative AI, including transformers, generative adversarial networks, and variational autoencoders. The generation process was guided by rule-based and physics- and chemistry-informed models for automatic labeling, with feedback loops to reduce invalid SMILES strings. The approach achieved 70 % molecular validity and 94 % novelty in generating new solvents for CO₂ capture and lignin applications. Model training averaged under one hour, and molecule generation took only seconds, significantly faster than traditional methods. Ensemble machine learning models assessed the environmental sustainability of candidates, and retrosynthesis analysis identified feasible, green synthesis pathways. This flexible, scalable methodology extends beyond solvent discovery to broader applications in process design and optimization, enabling the rapid generation of novel and cost-effective process configurations.
{"title":"Accelerated green material and solvent discovery with chemistry- and physics-guided generative AI","authors":"Eslam G. Al-Sakkari , Ahmed Ragab , Marzouk Benali , Olumoye Ajao , Daria C. Boffito , Hanane Dagdougui","doi":"10.1016/j.aichem.2025.100106","DOIUrl":"10.1016/j.aichem.2025.100106","url":null,"abstract":"<div><div>Carbon capture, utilization and storage (CCUS), along with lignocellulosic biomass valorization (e.g., lignin, cellulose), are promising decarbonization strategies for hard-to-abate industries. Green solvents, such as deep eutectic solvents and ionic liquids, enable efficient CO₂ capture and selective lignin extraction, enhancing lignin depolymerization into high-value products. However, current molecular design tools are slow and computationally expensive, limiting green material innovation. This study introduces a novel data-driven framework for green material discovery using generative AI, including transformers, generative adversarial networks, and variational autoencoders. The generation process was guided by rule-based and physics- and chemistry-informed models for automatic labeling, with feedback loops to reduce invalid SMILES strings. The approach achieved 70 % molecular validity and 94 % novelty in generating new solvents for CO₂ capture and lignin applications. Model training averaged under one hour, and molecule generation took only seconds, significantly faster than traditional methods. Ensemble machine learning models assessed the environmental sustainability of candidates, and retrosynthesis analysis identified feasible, green synthesis pathways. This flexible, scalable methodology extends beyond solvent discovery to broader applications in process design and optimization, enabling the rapid generation of novel and cost-effective process configurations.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"4 1","pages":"Article 100106"},"PeriodicalIF":0.0,"publicationDate":"2026-01-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925647","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-23DOI: 10.1016/j.aichem.2025.100105
Aaryashree , Arti Devi
The integration of machine learning (ML) with electrochemical sensors is transforming food safety and quality assessment by enabling quick, affordable, and highly sensitive detection of contaminants, adulterants, and spoilage indicators. Traditional electrochemical analysis faces challenges such as overlapping signals, nonlinear sensor responses, and matrix effects, which diminish accuracy and scalability. ML algorithms offer advanced data processing, feature extraction, and predictive modeling, significantly enhancing detection sensitivity, classification accuracy, and supporting real-time decision-making. This review explores the combined use of ML and electrochemical sensing in food analysis, focusing on key areas like pesticide and heavy metal detection, food authentication, shelf-life prediction, and microbial safety monitoring. It provides a comprehensive range of ML techniques, from basic algorithms like Support Vector Machines and Random Forests to advanced deep learning architectures, including Convolutional Neural Networks, Transformers, and Graph Neural Networks. Additionally, it highlights innovative applications and addresses critical challenges in real-world deployment, such as data scarcity, model generalizability, and the “black box” problem of interpretability. Strategies such as data augmentation, transfer learning, and explainable AI (XAI) are emerging as crucial solutions to enhance data availability and model transparency. The field is also advancing toward adaptive learning frameworks and integration with the Internet of Things (IoT), enabling continuous, networked monitoring throughout the food supply chain. By emphasizing both technical innovations and practical challenges, this review offers a solid foundation for researchers and professionals working at the intersection of electrochemical sensing, machine learning, and food safety analytics.
{"title":"Integrating machine learning with electrochemical sensors for intelligent food safety monitoring","authors":"Aaryashree , Arti Devi","doi":"10.1016/j.aichem.2025.100105","DOIUrl":"10.1016/j.aichem.2025.100105","url":null,"abstract":"<div><div>The integration of machine learning (ML) with electrochemical sensors is transforming food safety and quality assessment by enabling quick, affordable, and highly sensitive detection of contaminants, adulterants, and spoilage indicators. Traditional electrochemical analysis faces challenges such as overlapping signals, nonlinear sensor responses, and matrix effects, which diminish accuracy and scalability. ML algorithms offer advanced data processing, feature extraction, and predictive modeling, significantly enhancing detection sensitivity, classification accuracy, and supporting real-time decision-making. This review explores the combined use of ML and electrochemical sensing in food analysis, focusing on key areas like pesticide and heavy metal detection, food authentication, shelf-life prediction, and microbial safety monitoring. It provides a comprehensive range of ML techniques, from basic algorithms like Support Vector Machines and Random Forests to advanced deep learning architectures, including Convolutional Neural Networks, Transformers, and Graph Neural Networks. Additionally, it highlights innovative applications and addresses critical challenges in real-world deployment, such as data scarcity, model generalizability, and the “black box” problem of interpretability. Strategies such as data augmentation, transfer learning, and explainable AI (XAI) are emerging as crucial solutions to enhance data availability and model transparency. The field is also advancing toward adaptive learning frameworks and integration with the Internet of Things (IoT), enabling continuous, networked monitoring throughout the food supply chain. By emphasizing both technical innovations and practical challenges, this review offers a solid foundation for researchers and professionals working at the intersection of electrochemical sensing, machine learning, and food safety analytics.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"4 1","pages":"Article 100105"},"PeriodicalIF":0.0,"publicationDate":"2025-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884196","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-19DOI: 10.1016/j.aichem.2025.100104
Salman Khan , Nisar Ahmad , Sami Ullah , Liaqat Ali , Sajjad Ahmad , Hina Fazal
Border Disease Virus (BDV), a Flaviviridae pestivirus, cause major reproductive and financial losses to small ruminants, and no licensed vaccine is currently available. In this study, a multi-epitope vaccine (MEV) against BDV was designed using immunoinformatics approach. The construct exhibited favorable physiochemical properties, including an aliphatic index of 68.02, solubility probability of 0.96, and overall stability. It contained multiple high-scoring linear and conformational B-cell epitopes and showed strong predicted binding to MHC class I/II molecules. Molecular docking with TLR-4 revealed stable interactions (binding score: − 312.73). Immune simulations indicated robust primary IgM and secondary IgG responses with memory B- and T-cell formation. Codon optimization confirmed high expression potential in E. coli, (CAI: 1.0; GC content: 61 %), and in-silico cloning indicate vector compatibility. These results suggest that the proposed MEV has potential to induce both humoral and cellular immunity. Further experimental validation is recommended to confirm safety, immunogenicity, and protective efficacy.
{"title":"Computational design and immunoinformatic evaluation of a multi-epitope vaccine candidate against Border Disease Virus","authors":"Salman Khan , Nisar Ahmad , Sami Ullah , Liaqat Ali , Sajjad Ahmad , Hina Fazal","doi":"10.1016/j.aichem.2025.100104","DOIUrl":"10.1016/j.aichem.2025.100104","url":null,"abstract":"<div><div>Border Disease Virus (BDV), a Flaviviridae pestivirus, cause major reproductive and financial losses to small ruminants, and no licensed vaccine is currently available. In this study, a multi-epitope vaccine (MEV) against BDV was designed using immunoinformatics approach. The construct exhibited favorable physiochemical properties, including an aliphatic index of 68.02, solubility probability of 0.96, and overall stability. It contained multiple high-scoring linear and conformational B-cell epitopes and showed strong predicted binding to MHC class I/II molecules. Molecular docking with TLR-4 revealed stable interactions (binding score: − 312.73). Immune simulations indicated robust primary IgM and secondary IgG responses with memory B- and T-cell formation. Codon optimization confirmed high expression potential in <em>E. coli</em>, (CAI: 1.0; GC content: 61 %), and in-silico cloning indicate vector compatibility. These results suggest that the proposed MEV has potential to induce both humoral and cellular immunity. Further experimental validation is recommended to confirm safety, immunogenicity, and protective efficacy.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"4 1","pages":"Article 100104"},"PeriodicalIF":0.0,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145840726","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-17DOI: 10.1016/j.aichem.2025.100103
Sarmad Waleed , Shams ul Islam , Muhammad Saleem , Ali Ahmed
Background:
Accurate prediction of molecular properties is essential for accelerating drug discovery. While graph neural networks (GNNs) have emerged as a powerful tool for this task, they have not been systematically benchmarked against traditional machine learning methods, particularly regarding the crucial aspects of predictive accuracy, interpretability, and uncertainty.
Objective:
To systematically evaluate state-of-the-art GNN architectures against classical machine learning methods for predicting key physicochemical properties. This study provides a multi-faceted comparison of model performance, statistical robustness, prediction uncertainty, and chemical interpretability.
Methods:
We implemented and compared seven models: Graph Attention Networks (GAT), Graph Convolutional Networks (GCN), Graph Isomorphism Networks (GIN), SimpleGNN, Support Vector Regression (SVR), Random Forest, and ElasticNet. These were evaluated across three MoleculeNet datasets: ESOL (aqueous solubility), FreeSolv (hydration free energy), and Lipophilicity (partition coefficient). The evaluation framework included rigorous statistical testing, bootstrap-based uncertainty quantification, and analysis of GAT attention mechanisms for chemical insight.
Results:
GAT consistently achieved superior performance, with test RMSE values of 0.1863 (ESOL), 0.1953 (FreeSolv), and 0.4922 (Lipophilicity), outperforming traditional methods by a significant margin. GNNs demonstrated substantial advantages over classical approaches, which showed considerably higher prediction errors. Furthermore, GAT provided the most reliable predictions with the lowest uncertainty and generated chemically relevant insights through its attention mechanism, successfully identifying key functional groups driving molecular properties.
Conclusions:
This systematic evaluation provides compelling evidence for the superiority of GNNs, particularly GAT, over traditional machine learning for molecular property prediction. GAT’s high accuracy, combined with its robust uncertainty quantification and chemical interpretability, establishes it as a preferred computational approach for pharmaceutical research and development.
{"title":"Statistical comparison and uncertainty analysis of graph neural networks and machine learning models for molecular property prediction in drug discovery","authors":"Sarmad Waleed , Shams ul Islam , Muhammad Saleem , Ali Ahmed","doi":"10.1016/j.aichem.2025.100103","DOIUrl":"10.1016/j.aichem.2025.100103","url":null,"abstract":"<div><h3>Background:</h3><div>Accurate prediction of molecular properties is essential for accelerating drug discovery. While graph neural networks (GNNs) have emerged as a powerful tool for this task, they have not been systematically benchmarked against traditional machine learning methods, particularly regarding the crucial aspects of predictive accuracy, interpretability, and uncertainty.</div></div><div><h3>Objective:</h3><div>To systematically evaluate state-of-the-art GNN architectures against classical machine learning methods for predicting key physicochemical properties. This study provides a multi-faceted comparison of model performance, statistical robustness, prediction uncertainty, and chemical interpretability.</div></div><div><h3>Methods:</h3><div>We implemented and compared seven models: Graph Attention Networks (GAT), Graph Convolutional Networks (GCN), Graph Isomorphism Networks (GIN), SimpleGNN, Support Vector Regression (SVR), Random Forest, and ElasticNet. These were evaluated across three MoleculeNet datasets: ESOL (aqueous solubility), FreeSolv (hydration free energy), and Lipophilicity (partition coefficient). The evaluation framework included rigorous statistical testing, bootstrap-based uncertainty quantification, and analysis of GAT attention mechanisms for chemical insight.</div></div><div><h3>Results:</h3><div>GAT consistently achieved superior performance, with test RMSE values of 0.1863 (ESOL), 0.1953 (FreeSolv), and 0.4922 (Lipophilicity), outperforming traditional methods by a significant margin. GNNs demonstrated substantial advantages over classical approaches, which showed considerably higher prediction errors. Furthermore, GAT provided the most reliable predictions with the lowest uncertainty and generated chemically relevant insights through its attention mechanism, successfully identifying key functional groups driving molecular properties.</div></div><div><h3>Conclusions:</h3><div>This systematic evaluation provides compelling evidence for the superiority of GNNs, particularly GAT, over traditional machine learning for molecular property prediction. GAT’s high accuracy, combined with its robust uncertainty quantification and chemical interpretability, establishes it as a preferred computational approach for pharmaceutical research and development.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"4 1","pages":"Article 100103"},"PeriodicalIF":0.0,"publicationDate":"2025-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145798022","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper presents an innovative approach to the interpretable detection of organic pollutants dissolved in water – carbendazim, thiacloprid, and acetamiprid – by leveraging self-attention mechanisms to within the deep neural network (DNN) used for detection. The core contribution of our work is demonstrating how attention mechanisms can significantly enhance the interpretability and performance of pollutant detection in water using colloidal Surface-Enhanced Raman Spectroscopy (SERS) measurements. The cornerstone of our methodology is the optimization of the measurement process, aimed not merely at acquiring high-quality signals but at securing a high volume of data that embodies the full spectrum of measurement variability.
This optimization includes the development of a measurement protocol that involves (i) the fabrication of colloidal silver nanoparticles utilizing the method proposed by Leopold and Lendl, (ii) the aging of the colloidal mixture with the analytes for a predetermined period, and (iii) the SERS measurement settings. Each step is carefully calibrated to maximize the SERS response sensitivity and reproducibility for the detection of the targeted analytes. Building upon this optimized measurement framework, the paper introduces a deep learning algorithm with an embedded attention mechanism designed to focus on the most relevant spectral features for pollutant detection. Unlike traditional machine learning methods, which often lack interpretability, the proposed attention model provides clear insights into which features are deemed most important for the detection task, thereby offering a direct interpretation of the decision-making process of the neural network.
{"title":"Deep attention for interpretable detection of organic pollutants in water using colloidal SERS","authors":"Anvar Kunanbayev , Hirotsugu Hiramatsu , Wei-Liang Chen , Yu-Chun Huang , Yu-Ming Chang , Stefano Rini","doi":"10.1016/j.aichem.2025.100102","DOIUrl":"10.1016/j.aichem.2025.100102","url":null,"abstract":"<div><div>This paper presents an innovative approach to the interpretable detection of organic pollutants dissolved in water – carbendazim, thiacloprid, and acetamiprid – by leveraging self-attention mechanisms to within the deep neural network (DNN) used for detection. The core contribution of our work is demonstrating how attention mechanisms can significantly enhance the interpretability and performance of pollutant detection in water using colloidal Surface-Enhanced Raman Spectroscopy (SERS) measurements. The cornerstone of our methodology is the optimization of the measurement process, aimed not merely at acquiring high-quality signals but at securing a high volume of data that embodies the full spectrum of measurement variability.</div><div>This optimization includes the development of a measurement protocol that involves (i) the fabrication of colloidal silver nanoparticles utilizing the method proposed by Leopold and Lendl, (ii) the aging of the colloidal mixture with the analytes for a predetermined period, and (iii) the SERS measurement settings. Each step is carefully calibrated to maximize the SERS response sensitivity and reproducibility for the detection of the targeted analytes. Building upon this optimized measurement framework, the paper introduces a deep learning algorithm with an embedded attention mechanism designed to focus on the most relevant spectral features for pollutant detection. Unlike traditional machine learning methods, which often lack interpretability, the proposed attention model provides clear insights into which features are deemed most important for the detection task, thereby offering a direct interpretation of the decision-making process of the neural network.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"4 1","pages":"Article 100102"},"PeriodicalIF":0.0,"publicationDate":"2025-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145798021","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01DOI: 10.1016/j.aichem.2025.100101
Juda Baikété , Alhadji Malloum , Jeanet Conradie
Machine learning (ML) has become a powerful tool for predicting molecular physicochemical properties. It finds applications in various research and development sectors, such as materials science, pharmaceutical chemistry, and environmental science. However, systematic comparisons between different types of properties remain limited. In this study, we developed two structured datasets: a logP dataset containing 1117 molecules and a pKb dataset containing 1268 molecules. For logP, each molecule is represented by 623 molecular descriptors generated exclusively by RDKit/Mordred, while a combination of 150 quantum chemistry descriptors from DFT calculations and molecular fingerprints derived from RDKit is used for pKb. Several ML algorithms were evaluated using an identical workflow, and the relevance of the descriptors was analyzed using SHAP, followed by feature pruning based on correlation. For the logP dataset, the LightGradBoost model achieved an of 0.94, an RMSE of 0.31, and an MAE of 0.42 on the independent test set, accurately reproducing experimental logP values in the range of −11.6 to 1.58. For pKb prediction, Random Forest (RF) proved most accurate, with an MAE of 1.69 and an RMSE of 1.68, with predicted values covering the entire range of experimental pKb values (−37 to 29.2). Our results indicate that, while RDKit/Mordred descriptors can predict logP with high accuracy, pKb remains a more challenging property to model, even when incorporating high-level DFT descriptors. The study therefore proposes a unified framework for the comparative evaluation of cross-property machine learning models and highlights the influence of the type of descriptor and the choice of algorithm on performance for chemically distinct properties.
{"title":"Comparative study of machine learning methods for accurate prediction of logP and pKb","authors":"Juda Baikété , Alhadji Malloum , Jeanet Conradie","doi":"10.1016/j.aichem.2025.100101","DOIUrl":"10.1016/j.aichem.2025.100101","url":null,"abstract":"<div><div>Machine learning (ML) has become a powerful tool for predicting molecular physicochemical properties. It finds applications in various research and development sectors, such as materials science, pharmaceutical chemistry, and environmental science. However, systematic comparisons between different types of properties remain limited. In this study, we developed two structured datasets: a logP dataset containing 1117 molecules and a pKb dataset containing 1268 molecules. For logP, each molecule is represented by 623 molecular descriptors generated exclusively by RDKit/Mordred, while a combination of 150 quantum chemistry descriptors from DFT calculations and molecular fingerprints derived from RDKit is used for pKb. Several ML algorithms were evaluated using an identical workflow, and the relevance of the descriptors was analyzed using SHAP, followed by feature pruning based on correlation. For the logP dataset, the LightGradBoost model achieved an <span><math><msup><mrow><mi>R</mi></mrow><mrow><mn>2</mn></mrow></msup></math></span> of 0.94, an RMSE of 0.31, and an MAE of 0.42 on the independent test set, accurately reproducing experimental logP values in the range of −11.6 to 1.58. For pKb prediction, Random Forest (RF) proved most accurate, with an MAE of 1.69 and an RMSE of 1.68, with predicted values covering the entire range of experimental pKb values (−37 to 29.2). Our results indicate that, while RDKit/Mordred descriptors can predict logP with high accuracy, pKb remains a more challenging property to model, even when incorporating high-level DFT descriptors. The study therefore proposes a unified framework for the comparative evaluation of cross-property machine learning models and highlights the influence of the type of descriptor and the choice of algorithm on performance for chemically distinct properties.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"4 1","pages":"Article 100101"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145651810","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Elucidating the ways how the frequencies of vibrational modes are modulated by interactions with other moieties in a system is important for better utilization of the modes as probes to the structures and dynamics of condensed-phase systems. Here, such an analysis is carried out for the hydration-induced frequency shifts of the NH stretching mode of peptide. It is shown that, with a help of machine learning, it is possible to describe the frequency shifts as maps employing the electric fields operating on the atomic sites (H, N, C, and O) as descriptors. By comparing the results obtained for different combinations of descriptor sets and machine learning models, suitable ways to construct the maps of good performance are clarified. The nature of the electrostatic response of this mode, especially how the electric fields on the four atomic sites are involved in controlling the frequency shifts, is discussed.
{"title":"Electrostatic map for the hydration-induced frequency shifts of the NH stretching mode of peptide constructed using some machine learning models","authors":"Hajime Torii , Mikito Tsujimoto , Yukichi Kitamura","doi":"10.1016/j.aichem.2025.100098","DOIUrl":"10.1016/j.aichem.2025.100098","url":null,"abstract":"<div><div>Elucidating the ways how the frequencies of vibrational modes are modulated by interactions with other moieties in a system is important for better utilization of the modes as probes to the structures and dynamics of condensed-phase systems. Here, such an analysis is carried out for the hydration-induced frequency shifts of the NH stretching mode of peptide. It is shown that, with a help of machine learning, it is possible to describe the frequency shifts as maps employing the electric fields operating on the atomic sites (H, N, C, and O) as descriptors. By comparing the results obtained for different combinations of descriptor sets and machine learning models, suitable ways to construct the maps of good performance are clarified. The nature of the electrostatic response of this mode, especially how the electric fields on the four atomic sites are involved in controlling the frequency shifts, is discussed.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"3 2","pages":"Article 100098"},"PeriodicalIF":0.0,"publicationDate":"2025-11-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145579145","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-11-14DOI: 10.1016/j.aichem.2025.100099
Ye min Thant , Methawee Nukunudompanich , Chu-Chen Chueh , Manabu Ihara , Sergei Manzhos
Dedicated analog neurocomputing circuits are promising for high-throughput, low power consumption applications of machine learning (ML) and for applications where implementing a digital computer is unwieldy (remote locations; small, mobile, and autonomous devices, extreme conditions, etc.). Neural networks (NN) implemented in such circuits, however, must contend with circuit noise and the non-uniform shapes of the neuron activation function (NAF) due to the dispersion of performance characteristics of circuit elements (such as transistors or diodes implementing the neurons). We present a computational study of the impact of circuit noise and NAF inhomogeneity in regression problems as a function of NN architecture and training regimes. We focus on one application that requires high-throughput ML: materials informatics, using as representative problem ML of formation energies vs. lowest-energy isomer of peri-condensed hydrocarbons, formation energies and band gaps of double perovskites, and zero point vibrational energies of molecules from QM9 dataset. We show that in these applications, NNs generally possess low noise tolerance with the model accuracy rapidly degrading with noise level. Single-hidden layer NNs, and NNs with larger-than-optimal sizes are somewhat more noise-tolerant. Models that show less overfitting (not necessarily the lowest test set error) are more noise-tolerant. Importantly, we demonstrate that the effect of activation function inhomogeneity can be palliated by retraining the NN using practically realized shapes of NAFs.
{"title":"Neural networks for neurocomputing circuits: A computational study of tolerance to noise and activation function non-uniformity when machine learning materials properties","authors":"Ye min Thant , Methawee Nukunudompanich , Chu-Chen Chueh , Manabu Ihara , Sergei Manzhos","doi":"10.1016/j.aichem.2025.100099","DOIUrl":"10.1016/j.aichem.2025.100099","url":null,"abstract":"<div><div>Dedicated analog neurocomputing circuits are promising for high-throughput, low power consumption applications of machine learning (ML) and for applications where implementing a digital computer is unwieldy (remote locations; small, mobile, and autonomous devices, extreme conditions, etc.). Neural networks (NN) implemented in such circuits, however, must contend with circuit noise and the non-uniform shapes of the neuron activation function (NAF) due to the dispersion of performance characteristics of circuit elements (such as transistors or diodes implementing the neurons). We present a computational study of the impact of circuit noise and NAF inhomogeneity in regression problems as a function of NN architecture and training regimes. We focus on one application that requires high-throughput ML: materials informatics, using as representative problem ML of formation energies vs. lowest-energy isomer of peri-condensed hydrocarbons, formation energies and band gaps of double perovskites, and zero point vibrational energies of molecules from QM9 dataset. We show that in these applications, NNs generally possess low noise tolerance with the model accuracy rapidly degrading with noise level. Single-hidden layer NNs, and NNs with larger-than-optimal sizes are somewhat more noise-tolerant. Models that show less overfitting (not necessarily the lowest test set error) are more noise-tolerant. Importantly, we demonstrate that the effect of activation function inhomogeneity can be palliated by retraining the NN using practically realized shapes of NAFs.</div></div>","PeriodicalId":72302,"journal":{"name":"Artificial intelligence chemistry","volume":"3 2","pages":"Article 100099"},"PeriodicalIF":0.0,"publicationDate":"2025-11-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145527993","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号