Pub Date : 2025-06-17DOI: 10.1186/s13321-025-00958-w
Nico Domschke, Bruno J. Schmidt, Thomas Gatter, Richard Golnik, Paul Eisenhuth, Fabian Liessmann, Jens Meiler, Peter F. Stadler
Genetic algorithms are a powerful method to solve optimization problems with complex cost functions over vast search spaces that rely in particular on recombining parts of previous solutions. Crossover operators play a crucial role in this context. Here, we describe a large class of these operators designed for searching over spaces of graphs. These operators are based on introducing small cuts into graphs and rejoining the resulting induced subgraphs of two parents. This form of cut-and-join crossover can be restricted in a consistent way to preserve local properties such as vertex-degrees (valency), or bond-orders, as well as global properties such as graph-theoretic planarity. In contrast to crossover on strings, cut-and-join crossover on graphs is powerful enough to ergodically explore chemical space even in the absence of mutation operators. Extensive benchmarking shows that the offspring of molecular graphs are again plausible molecules with high probability, while at the same time crossover drastically increases the diversity compared to initial molecule libraries. Moreover, desirable properties such as favorable indices of synthesizability are preserved with sufficient frequency that candidate offsprings can be filtered efficiently for such properties. As an application we utilized the cut-and-join crossover in REvoLd, a GA-based system for computer-aided drug design. In optimization runs searching for ligands binding to four different target proteins we consistently found candidate molecules with binding constants exceeding the best known binders as well as candidates found in make-on-demand libraries. Scientific contribution We define cut-and-join crossover operators on a variety of graph classes including molecular graphs. This constitutes a mathematically simple and well-characterized approach to recombination of molecules that performed very well in real-life CADD tasks.
{"title":"Crossover operators for molecular graphs with an application to virtual drug screening","authors":"Nico Domschke, Bruno J. Schmidt, Thomas Gatter, Richard Golnik, Paul Eisenhuth, Fabian Liessmann, Jens Meiler, Peter F. Stadler","doi":"10.1186/s13321-025-00958-w","DOIUrl":"https://doi.org/10.1186/s13321-025-00958-w","url":null,"abstract":"Genetic algorithms are a powerful method to solve optimization problems with complex cost functions over vast search spaces that rely in particular on recombining parts of previous solutions. Crossover operators play a crucial role in this context. Here, we describe a large class of these operators designed for searching over spaces of graphs. These operators are based on introducing small cuts into graphs and rejoining the resulting induced subgraphs of two parents. This form of cut-and-join crossover can be restricted in a consistent way to preserve local properties such as vertex-degrees (valency), or bond-orders, as well as global properties such as graph-theoretic planarity. In contrast to crossover on strings, cut-and-join crossover on graphs is powerful enough to ergodically explore chemical space even in the absence of mutation operators. Extensive benchmarking shows that the offspring of molecular graphs are again plausible molecules with high probability, while at the same time crossover drastically increases the diversity compared to initial molecule libraries. Moreover, desirable properties such as favorable indices of synthesizability are preserved with sufficient frequency that candidate offsprings can be filtered efficiently for such properties. As an application we utilized the cut-and-join crossover in REvoLd, a GA-based system for computer-aided drug design. In optimization runs searching for ligands binding to four different target proteins we consistently found candidate molecules with binding constants exceeding the best known binders as well as candidates found in make-on-demand libraries. Scientific contribution We define cut-and-join crossover operators on a variety of graph classes including molecular graphs. This constitutes a mathematically simple and well-characterized approach to recombination of molecules that performed very well in real-life CADD tasks.","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"44 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-06-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144311943","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-16DOI: 10.1186/s13321-025-01033-0
Raheel Hammad, Sownyak Mondal
The Gibbs free energy of an inorganic material represents its maximum reversible work potential under constant temperature and pressure. Its calculation is crucial for understanding material stability, phase transitions, and chemical reactions, thus guiding optimization for diverse applications like catalysis and energy storage. In this study, we have developed a Physics-Informed Neural Network model that leverages the Gibbs free energy equation. The overall loss function is adjusted to allow the model to simultaneously predict all three thermodynamic quantities, including Gibbs free energy, total energy, and entropy, thus transforming it into a multi-output model. In recent literature, there is a growing emphasis on evaluating machine learning models under challenging conditions, such as small datasets and out-of-distribution predictions. Reflecting this trend, we have rigorously benchmarked our model across these scenarios, demonstrating its robustness and adaptability. It turns out that our model demonstrates a 43% improvement for normal scenario and even more in out-of-distribution regime compared to the next-best model. Scientific Contribution This study introduces the application of a Physics-Informed Neural Network to simultaneously compute multiple thermodynamic properties, including Gibbs free energy, total energy, and entropy. By integrating the Gibbs free energy equation into the loss function, the model achieves superior accuracy in low data regimes and enhances robustness in the out-of-distribution scenarios.
{"title":"Advancements in thermochemical predictions: a multi-output thermodynamics-informed neural network approach","authors":"Raheel Hammad, Sownyak Mondal","doi":"10.1186/s13321-025-01033-0","DOIUrl":"https://doi.org/10.1186/s13321-025-01033-0","url":null,"abstract":"The Gibbs free energy of an inorganic material represents its maximum reversible work potential under constant temperature and pressure. Its calculation is crucial for understanding material stability, phase transitions, and chemical reactions, thus guiding optimization for diverse applications like catalysis and energy storage. In this study, we have developed a Physics-Informed Neural Network model that leverages the Gibbs free energy equation. The overall loss function is adjusted to allow the model to simultaneously predict all three thermodynamic quantities, including Gibbs free energy, total energy, and entropy, thus transforming it into a multi-output model. In recent literature, there is a growing emphasis on evaluating machine learning models under challenging conditions, such as small datasets and out-of-distribution predictions. Reflecting this trend, we have rigorously benchmarked our model across these scenarios, demonstrating its robustness and adaptability. It turns out that our model demonstrates a 43% improvement for normal scenario and even more in out-of-distribution regime compared to the next-best model. Scientific Contribution This study introduces the application of a Physics-Informed Neural Network to simultaneously compute multiple thermodynamic properties, including Gibbs free energy, total energy, and entropy. By integrating the Gibbs free energy equation into the loss function, the model achieves superior accuracy in low data regimes and enhances robustness in the out-of-distribution scenarios.","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"11 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144296245","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-16DOI: 10.1186/s13321-025-01040-1
Palistha Shrestha, Chandana S. Talwar, Jeevan Kandel, Kwang-Hyun Park, Kil To Chong, Eui-Jeon Woo, Hilal Tayara
Nanobodies offer significant therapeutic potential due to their small size, stability, and versatility. Although advancements in computational protein design have made designing de novo nanobodies increasingly feasible, there are limited tools specifically tailored for this purpose. Rosetta with its specialized protocols, is a prominent tool for nanobody design but is limited by a high false-negative rate, necessitating extensive high-throughput screening. This results in increased costs, time, and labor due to the need for large-scale experimentation and detailed structural analysis. To address current challenges in nanobody design, we introduce NanoBinder, an interpretable machine learning model that predicts nanobody-antigen binding using Rosetta energy scores. NanoBinder utilizes a Random Forest model trained on experimentally validated complexes and can be seamlessly integrated into the Rosetta software. It employs SHAP summary plots for interpretability, which helps identify key features influencing binding interactions. Experimentally validated on forty-nine diverse nanobodies, NanoBinder accurately predicts non-binders and shows reasonable performance in identifying binders. This approach significantly enhances predictive accuracy, reduces the need for extensive experimental assays, and accelerates nanobody development, thereby offering a powerful tool to mitigate the costs, time, and labor associated with high-throughput screening. Scientific contribution This study introduces NanoBinder, a machine learning framework for predicting nanobody-antigen binding using Rosetta-derived energy features. Through rigorous experimental validation across diverse nanobody sets, NanoBinder enhances nanobody screening workflows by reducing false positives and minimizing reliance on extensive wet-lab assays. The approach bridges the gap between physics-based modeling and data-driven prediction in nanobody design.
{"title":"NanoBinder: a machine learning assisted nanobody binding prediction tool using Rosetta energy scores","authors":"Palistha Shrestha, Chandana S. Talwar, Jeevan Kandel, Kwang-Hyun Park, Kil To Chong, Eui-Jeon Woo, Hilal Tayara","doi":"10.1186/s13321-025-01040-1","DOIUrl":"https://doi.org/10.1186/s13321-025-01040-1","url":null,"abstract":"Nanobodies offer significant therapeutic potential due to their small size, stability, and versatility. Although advancements in computational protein design have made designing de novo nanobodies increasingly feasible, there are limited tools specifically tailored for this purpose. Rosetta with its specialized protocols, is a prominent tool for nanobody design but is limited by a high false-negative rate, necessitating extensive high-throughput screening. This results in increased costs, time, and labor due to the need for large-scale experimentation and detailed structural analysis. To address current challenges in nanobody design, we introduce NanoBinder, an interpretable machine learning model that predicts nanobody-antigen binding using Rosetta energy scores. NanoBinder utilizes a Random Forest model trained on experimentally validated complexes and can be seamlessly integrated into the Rosetta software. It employs SHAP summary plots for interpretability, which helps identify key features influencing binding interactions. Experimentally validated on forty-nine diverse nanobodies, NanoBinder accurately predicts non-binders and shows reasonable performance in identifying binders. This approach significantly enhances predictive accuracy, reduces the need for extensive experimental assays, and accelerates nanobody development, thereby offering a powerful tool to mitigate the costs, time, and labor associated with high-throughput screening. Scientific contribution This study introduces NanoBinder, a machine learning framework for predicting nanobody-antigen binding using Rosetta-derived energy features. Through rigorous experimental validation across diverse nanobody sets, NanoBinder enhances nanobody screening workflows by reducing false positives and minimizing reliance on extensive wet-lab assays. The approach bridges the gap between physics-based modeling and data-driven prediction in nanobody design.","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"227 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-06-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144296244","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-10DOI: 10.1186/s13321-025-01039-8
Qianrong Guo, Saiveth Hernandez-Hernandez, Pedro J. Ballester
Virtual Screening (VS) of large compound libraries using Artificial Intelligence (AI) models is a highly effective approach for early drug discovery. Data splitting is crucial for benchmarking the performance of such AI models. Traditional random data splits often result in structurally similar molecules in both training and test sets, which conflict with the reality of VS libraries that typically contain structurally diverse compounds. To tackle this challenge, scaffold split, which groups molecules by shared core structure, and Butina clustering, which clusters molecules by chemotypes, have long been used. However, we show that these methods still introduce high similarities between training and test sets, leading to overestimated model performance. Our study examined four representative AI models across 60 NCI-60 datasets, each comprising approximately 33,000–54,000 molecules tested on different cancer cell lines. Each dataset was split in four ways: random, scaffold, Butina clustering and the more realistic Uniform Manifold Approximation and Projection (UMAP) clustering. Using Linear Regression, Random Forest, Transformer-CNN, and GEM, we trained a total of 8400 models and evaluated under four splitting methods. These comprehensive results show that UMAP split provides more challenging and realistic benchmarks for model evaluation, followed by Butina splits, then scaffold splits and closely after random splits. Consequently, we recommend using UMAP splits instead of overly optimistic Butina splits and especially scaffold splits for molecular property prediction, including VS. Lastly, we illustrate how misaligned ROC AUC is with VS goals, despite its common use. The code and datasets for reproducibility are available at https://github.com/Rong830/UMAP_split_for_VS and archived in https://zenodo.org/records/14736486 . Scientific contribution This work advances the field by introducing UMAP clustering as a robust splitting method for molecular datasets, improving over traditional methods like Butina clustering and especially scaffold splits. It offers a new evaluation framework to benchmark AI models under more realistic conditions, fostering progress in molecular property prediction. The findings also show how inappropriate the use of ROC AUC for virtual screening (VS) continues to be, despite its popularity, emphasizing the need for context-specific evaluation metrics.
{"title":"UMAP-based clustering split for rigorous evaluation of AI models for virtual screening on cancer cell lines*","authors":"Qianrong Guo, Saiveth Hernandez-Hernandez, Pedro J. Ballester","doi":"10.1186/s13321-025-01039-8","DOIUrl":"https://doi.org/10.1186/s13321-025-01039-8","url":null,"abstract":"Virtual Screening (VS) of large compound libraries using Artificial Intelligence (AI) models is a highly effective approach for early drug discovery. Data splitting is crucial for benchmarking the performance of such AI models. Traditional random data splits often result in structurally similar molecules in both training and test sets, which conflict with the reality of VS libraries that typically contain structurally diverse compounds. To tackle this challenge, scaffold split, which groups molecules by shared core structure, and Butina clustering, which clusters molecules by chemotypes, have long been used. However, we show that these methods still introduce high similarities between training and test sets, leading to overestimated model performance. Our study examined four representative AI models across 60 NCI-60 datasets, each comprising approximately 33,000–54,000 molecules tested on different cancer cell lines. Each dataset was split in four ways: random, scaffold, Butina clustering and the more realistic Uniform Manifold Approximation and Projection (UMAP) clustering. Using Linear Regression, Random Forest, Transformer-CNN, and GEM, we trained a total of 8400 models and evaluated under four splitting methods. These comprehensive results show that UMAP split provides more challenging and realistic benchmarks for model evaluation, followed by Butina splits, then scaffold splits and closely after random splits. Consequently, we recommend using UMAP splits instead of overly optimistic Butina splits and especially scaffold splits for molecular property prediction, including VS. Lastly, we illustrate how misaligned ROC AUC is with VS goals, despite its common use. The code and datasets for reproducibility are available at https://github.com/Rong830/UMAP_split_for_VS and archived in https://zenodo.org/records/14736486 . Scientific contribution This work advances the field by introducing UMAP clustering as a robust splitting method for molecular datasets, improving over traditional methods like Butina clustering and especially scaffold splits. It offers a new evaluation framework to benchmark AI models under more realistic conditions, fostering progress in molecular property prediction. The findings also show how inappropriate the use of ROC AUC for virtual screening (VS) continues to be, despite its popularity, emphasizing the need for context-specific evaluation metrics.","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"218 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-06-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144260195","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-04DOI: 10.1186/s13321-025-01035-y
Yongna Yuan, Xiaohang Pan, Xiaohong Li, Ruisheng Zhang, Wei Su
Deep generative models provide a powerful solution for the de novo design of molecules. However, the majority of existing methods only generate molecules for a single target. Generating molecules with biological activities against multiple specific targets and desired properties remains an extremely difficult challenge. In this study, we propose a novel 3D molecule generation framework based on reinforcement learning and diffusion model to generate molecules with predefined properties for given multiple targets. The proposed framework, MDRL, uses a diffusion model to understand the 3D chemical structure of molecules and employs Kolmogorov-Arnold Networks instead of Multilayer Perceptron to enhance model performance. Through reinforcement learning, the framework is able to generate molecules that simultaneously target two targets and further optimizes multiple molecular properties. Experimental results show that our model exhibits comparable performance to various state-of-the-art molecular generation models, and MDRL can effectively navigate chemical space to design polypharmacological compounds and control multiple molecular properties. In multiple case studies, we verify that the generated molecules can simultaneously target two targets through molecular docking and assess the model’s ability to control multiple molecular properties. The results in this study highlight the advantages and practicalities of our model in generating polypharmacological compounds with desired properties. This study introduces MDRL, a 3D molecular generation framework integrating diffusion models and reinforcement learning for joint optimization of multi-target binding and molecular properties. MDRL shows improvements over existing methods in controlling drug-relevant properties and enhancing multi-target affinity. Experimental results demonstrate that MDRL efficiently generates drug-like compounds with robust polypharmacological profiles, offering a novel strategy for multi-target drug design.
{"title":"A 3D generation framework using diffusion model and reinforcement learning to generate multi-target compounds with desired properties","authors":"Yongna Yuan, Xiaohang Pan, Xiaohong Li, Ruisheng Zhang, Wei Su","doi":"10.1186/s13321-025-01035-y","DOIUrl":"https://doi.org/10.1186/s13321-025-01035-y","url":null,"abstract":"Deep generative models provide a powerful solution for the de novo design of molecules. However, the majority of existing methods only generate molecules for a single target. Generating molecules with biological activities against multiple specific targets and desired properties remains an extremely difficult challenge. In this study, we propose a novel 3D molecule generation framework based on reinforcement learning and diffusion model to generate molecules with predefined properties for given multiple targets. The proposed framework, MDRL, uses a diffusion model to understand the 3D chemical structure of molecules and employs Kolmogorov-Arnold Networks instead of Multilayer Perceptron to enhance model performance. Through reinforcement learning, the framework is able to generate molecules that simultaneously target two targets and further optimizes multiple molecular properties. Experimental results show that our model exhibits comparable performance to various state-of-the-art molecular generation models, and MDRL can effectively navigate chemical space to design polypharmacological compounds and control multiple molecular properties. In multiple case studies, we verify that the generated molecules can simultaneously target two targets through molecular docking and assess the model’s ability to control multiple molecular properties. The results in this study highlight the advantages and practicalities of our model in generating polypharmacological compounds with desired properties. This study introduces MDRL, a 3D molecular generation framework integrating diffusion models and reinforcement learning for joint optimization of multi-target binding and molecular properties. MDRL shows improvements over existing methods in controlling drug-relevant properties and enhancing multi-target affinity. Experimental results demonstrate that MDRL efficiently generates drug-like compounds with robust polypharmacological profiles, offering a novel strategy for multi-target drug design.","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"16 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144211377","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-02DOI: 10.1186/s13321-025-01034-z
Lun Zhu, Qingchao Zhang, Sen Yang
Recent progress in computational biology has driven the development of machine learning models for predicting protein post-translational modification sites. However, challenges such as data imbalance and limited sequence-context representation continue to hinder prediction accuracy, particularly for less frequent modifications like succinylation. In this study, we propose RLSuccSite, a reinforcement learning-based framework specifically designed to predict succinylation sites by addressing the class imbalance issue via a dynamic with balanced reward mechanism. To enhance sequence feature representation, this study also introduces Three-Peaks Enhanced Method for Physicochemical Property Scores (TPEM-PPS), a physicochemical property-driven feature extraction method that incorporates position-aware scoring to reflect amino acid contributions more effectively. The code and data of RLSuccSite can be obtained from the website: https://github.com/Zhangqingchao-Ch/RLSuccSite.git . Scientific contribution This study applies reinforcement learning to protein succinylation sites prediction, introducing a dynamic with balanced reward mechanism that effectively addresses dataset imbalance. Additionally, this study proposes a novel Three-Peaks Enhanced Method for Physicochemical Scoring, which captures residue contributions with higher precision than traditional feature extraction techniques.
{"title":"RLSuccSite: succinylation sites prediction based on reinforcement learning dynamic with balanced reward mechanism and three-peaks enhanced method for physicochemical property scores","authors":"Lun Zhu, Qingchao Zhang, Sen Yang","doi":"10.1186/s13321-025-01034-z","DOIUrl":"https://doi.org/10.1186/s13321-025-01034-z","url":null,"abstract":"Recent progress in computational biology has driven the development of machine learning models for predicting protein post-translational modification sites. However, challenges such as data imbalance and limited sequence-context representation continue to hinder prediction accuracy, particularly for less frequent modifications like succinylation. In this study, we propose RLSuccSite, a reinforcement learning-based framework specifically designed to predict succinylation sites by addressing the class imbalance issue via a dynamic with balanced reward mechanism. To enhance sequence feature representation, this study also introduces Three-Peaks Enhanced Method for Physicochemical Property Scores (TPEM-PPS), a physicochemical property-driven feature extraction method that incorporates position-aware scoring to reflect amino acid contributions more effectively. The code and data of RLSuccSite can be obtained from the website: https://github.com/Zhangqingchao-Ch/RLSuccSite.git . Scientific contribution This study applies reinforcement learning to protein succinylation sites prediction, introducing a dynamic with balanced reward mechanism that effectively addresses dataset imbalance. Additionally, this study proposes a novel Three-Peaks Enhanced Method for Physicochemical Scoring, which captures residue contributions with higher precision than traditional feature extraction techniques. ","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"9 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-06-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144193336","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-31DOI: 10.1186/s13321-025-01025-0
Eduardo Illueca Fernández, Antonio Jesús Jara Valera, Jesualdo Tomás Fernández Breis
Persistent air quality pollution poses a serious threat to human health, and is one of the action points that policy makers should monitor according to the Directive 2008/50/EC. While deploying a massive network of hyperlocal sensors could provide extensive monitoring, this approach cannot generate geospatial continuous data and present several challenges in terms of logistics. Thus, developing accurate and trustable expert systems based on chemistry transport models is a key strategy for environmental protection. However, chemistry transport models present an important lack of standardization, and the formats are not interoperable between different systems, which limits the use for different stakeholders. In this context, semantic technologies provide methods and standards for scientific data and make information readable for expert systems. Therefore, this paper proposes a novel methodology for an ontology driven transformation for CHIMERE simulations, a chemistry transport model, allowing to generate knowledge graphs representing air quality information. It enables the transformation of netCDF files into RDF triples for short term air quality forecasting. Concretely, we utilize the Semantic Web Integration Tool (SWIT) framework for mapping individuals using an ontology as a template. Then, a new ontology for CHIMERE has been defined in this work, reusing concepts for other standards in the state of the art. Our approach demonstrates that RDF files can be created from netCDF in a linear computational time, allowing the scalability for expert systems. In addition, the ontology complains with the OQuaRE quality metrics and can be extended in future extensions to be applied to other chemistry transport models. Development of the first ontology for a chemistry transport model. FAIRification of physical models thanks to the generation of knowledge graphs from netCDF files. The ontology proposed is published in PURL ( https://purl.org/chimere-ontology ) and the knowledge graph generated for a 72-h simulation can be accessed in the following repository: https://doi.org/10.5281/zenodo.13981544 .
{"title":"Representation of chemistry transport models simulations using knowledge graphs","authors":"Eduardo Illueca Fernández, Antonio Jesús Jara Valera, Jesualdo Tomás Fernández Breis","doi":"10.1186/s13321-025-01025-0","DOIUrl":"https://doi.org/10.1186/s13321-025-01025-0","url":null,"abstract":"Persistent air quality pollution poses a serious threat to human health, and is one of the action points that policy makers should monitor according to the Directive 2008/50/EC. While deploying a massive network of hyperlocal sensors could provide extensive monitoring, this approach cannot generate geospatial continuous data and present several challenges in terms of logistics. Thus, developing accurate and trustable expert systems based on chemistry transport models is a key strategy for environmental protection. However, chemistry transport models present an important lack of standardization, and the formats are not interoperable between different systems, which limits the use for different stakeholders. In this context, semantic technologies provide methods and standards for scientific data and make information readable for expert systems. Therefore, this paper proposes a novel methodology for an ontology driven transformation for CHIMERE simulations, a chemistry transport model, allowing to generate knowledge graphs representing air quality information. It enables the transformation of netCDF files into RDF triples for short term air quality forecasting. Concretely, we utilize the Semantic Web Integration Tool (SWIT) framework for mapping individuals using an ontology as a template. Then, a new ontology for CHIMERE has been defined in this work, reusing concepts for other standards in the state of the art. Our approach demonstrates that RDF files can be created from netCDF in a linear computational time, allowing the scalability for expert systems. In addition, the ontology complains with the OQuaRE quality metrics and can be extended in future extensions to be applied to other chemistry transport models. Development of the first ontology for a chemistry transport model. FAIRification of physical models thanks to the generation of knowledge graphs from netCDF files. The ontology proposed is published in PURL ( https://purl.org/chimere-ontology ) and the knowledge graph generated for a 72-h simulation can be accessed in the following repository: https://doi.org/10.5281/zenodo.13981544 .","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"3 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144188999","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-31DOI: 10.1186/s13321-025-01036-x
Alexandre Varnek, Gilles Marcou, Dragos Horvath
While chemoinformatics is a well-established scientific field, its integration into university curricula is rarely discussed. In this work, we share our experience in developing a chemoinformatics curriculum at the University of Strasbourg and highlight the main challenges in higher education for this discipline.
{"title":"Higher education in chemoinformatics: achievements and challenges","authors":"Alexandre Varnek, Gilles Marcou, Dragos Horvath","doi":"10.1186/s13321-025-01036-x","DOIUrl":"https://doi.org/10.1186/s13321-025-01036-x","url":null,"abstract":"While chemoinformatics is a well-established scientific field, its integration into university curricula is rarely discussed. In this work, we share our experience in developing a chemoinformatics curriculum at the University of Strasbourg and highlight the main challenges in higher education for this discipline.","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"28 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144188912","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-31DOI: 10.1186/s13321-025-01028-x
Tuan Le, Julian Cremer, Djork-Arné Clevert, Kristof T. Schütt
We introduce PoLiGenX, a novel generative model for de novo ligand design that employs latent-conditioned, target-aware equivariant diffusion. Our approach leverages the conditioning of the ligand generation process on reference molecules located within a specific protein pocket. By doing so, PoLiGenX generates shape-similar ligands that are adapted to the target pocket, enabling effective applications in target-aware hit expansion and hit optimization. Our experimental results underscore the efficacy of PoLiGenX in advancing ligand design. Notably, docking analyses reveal that the ligands generated by PoLiGenX show enhanced binding affinities relative to their reference molecules, all while retaining a similar molecular shape, but also retaining better poses with lower strain energies and less steric clashes. Furthermore, the model promotes substantial chemical diversity, facilitating the exploration of broader and more varied chemical spaces. Importantly, the generated ligands were assessed for drug-likeness using Lipinski’s rule of five, demonstrating superior adherence to drug-likeness criteria compared to the reference dataset. This work represents a step forward in the controlled and precise generation of therapeutically relevant de novo ligands tailored for specific protein targets, contributing to progress in computational drug discovery and ligand design. We present a latent-conditioning method within diffusion models to enable the controllable generation of ligands in structure-based drug design that are similar to a reference ligand. We show that the generated ligands obtained via latent-conditioning achieve favorable ligand poses with reduced steric clashes and lower strain energies compared to diffusion models that only condition on the protein pocket alone. We demonstrate that the ligand generation can be further constrained using an importance sampling algorithm with external surrogate models that account for molecular properties such as synthetic accessibility.
{"title":"Equivariant diffusion for structure-based de novo ligand generation with latent-conditioning","authors":"Tuan Le, Julian Cremer, Djork-Arné Clevert, Kristof T. Schütt","doi":"10.1186/s13321-025-01028-x","DOIUrl":"https://doi.org/10.1186/s13321-025-01028-x","url":null,"abstract":"We introduce PoLiGenX, a novel generative model for de novo ligand design that employs latent-conditioned, target-aware equivariant diffusion. Our approach leverages the conditioning of the ligand generation process on reference molecules located within a specific protein pocket. By doing so, PoLiGenX generates shape-similar ligands that are adapted to the target pocket, enabling effective applications in target-aware hit expansion and hit optimization. Our experimental results underscore the efficacy of PoLiGenX in advancing ligand design. Notably, docking analyses reveal that the ligands generated by PoLiGenX show enhanced binding affinities relative to their reference molecules, all while retaining a similar molecular shape, but also retaining better poses with lower strain energies and less steric clashes. Furthermore, the model promotes substantial chemical diversity, facilitating the exploration of broader and more varied chemical spaces. Importantly, the generated ligands were assessed for drug-likeness using Lipinski’s rule of five, demonstrating superior adherence to drug-likeness criteria compared to the reference dataset. This work represents a step forward in the controlled and precise generation of therapeutically relevant de novo ligands tailored for specific protein targets, contributing to progress in computational drug discovery and ligand design. We present a latent-conditioning method within diffusion models to enable the controllable generation of ligands in structure-based drug design that are similar to a reference ligand. We show that the generated ligands obtained via latent-conditioning achieve favorable ligand poses with reduced steric clashes and lower strain energies compared to diffusion models that only condition on the protein pocket alone. We demonstrate that the ligand generation can be further constrained using an importance sampling algorithm with external surrogate models that account for molecular properties such as synthetic accessibility.","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"7 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144188881","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-05-31DOI: 10.1186/s13321-025-01029-w
Alicia Olivares-Gil, José A. Barbero-Aparicio, Juan J. Rodríguez, José F. Díez-Pastor, César García-Osorio, Mehdi D. Davari
Protein fitness prediction plays a crucial role in the advancement of protein engineering endeavours. However, the combinatorial complexity of the protein sequence space and the limited availability of assay-labelled data hinder the efficient optimization of protein properties. Data-driven strategies utilizing machine learning methods have emerged as a promising solution, yet their dependence on labelled training datasets poses a significant obstacle. To overcome this challenge, in this work, we explore various ways of introducing the latent information present in evolutionarily related sequences (homologous sequences) into the training process. To do so, we establish several strategies based on semi-supervised learning (unsupervised pre-processing and wrapper methods) and perform a comprehensive comparison using 19 datasets containing protein-fitness pairs. Our findings reveal that using the information present in the homologous sequences can improve the performance of the models, especially when the number of available labelled sequences is considerably low. Specifically, the combination of a sequence encoding method based on Direct Coupling Analysis (DCA), with MERGE (a hybrid regression framework that combines evolutionary information with supervised learning) and an SVM regressor, outperforms other encodings (PAM250, UniRep, eUniRep) and other semi-supervised wrapper methods (Tri-Training Regressor, Co-Training Regressor). In summary, the demonstrated performance gains of this strategy mark a substantial leap towards more robust and reliable predictive models for protein engineering tasks. This advancement holds the potential to streamline the design and optimisation of proteins for diverse applications in biotechnology and therapeutics. We explore several semi-supervised learning strategies capable of including the homologous sequences (unlabelled) to the protein of interest in the training process. Among them, we present two new methods to exploit the information in the homologous sequences: i) a new generalised version of MERGE capable of employing any regressor as a base estimator; ii) the Tri-Training Regressor method, an adaptation of the Tri-Training method for regression problems. We find that the information inherent in the homologous sequences has the ability to improve the predictive capacity of models when the number of available sequences is scarce, especially when using the DCA encoding together with MERGE and an SVM regressor.
{"title":"Semi-supervised prediction of protein fitness for data-driven protein engineering","authors":"Alicia Olivares-Gil, José A. Barbero-Aparicio, Juan J. Rodríguez, José F. Díez-Pastor, César García-Osorio, Mehdi D. Davari","doi":"10.1186/s13321-025-01029-w","DOIUrl":"https://doi.org/10.1186/s13321-025-01029-w","url":null,"abstract":"Protein fitness prediction plays a crucial role in the advancement of protein engineering endeavours. However, the combinatorial complexity of the protein sequence space and the limited availability of assay-labelled data hinder the efficient optimization of protein properties. Data-driven strategies utilizing machine learning methods have emerged as a promising solution, yet their dependence on labelled training datasets poses a significant obstacle. To overcome this challenge, in this work, we explore various ways of introducing the latent information present in evolutionarily related sequences (homologous sequences) into the training process. To do so, we establish several strategies based on semi-supervised learning (unsupervised pre-processing and wrapper methods) and perform a comprehensive comparison using 19 datasets containing protein-fitness pairs. Our findings reveal that using the information present in the homologous sequences can improve the performance of the models, especially when the number of available labelled sequences is considerably low. Specifically, the combination of a sequence encoding method based on Direct Coupling Analysis (DCA), with MERGE (a hybrid regression framework that combines evolutionary information with supervised learning) and an SVM regressor, outperforms other encodings (PAM250, UniRep, eUniRep) and other semi-supervised wrapper methods (Tri-Training Regressor, Co-Training Regressor). In summary, the demonstrated performance gains of this strategy mark a substantial leap towards more robust and reliable predictive models for protein engineering tasks. This advancement holds the potential to streamline the design and optimisation of proteins for diverse applications in biotechnology and therapeutics. We explore several semi-supervised learning strategies capable of including the homologous sequences (unlabelled) to the protein of interest in the training process. Among them, we present two new methods to exploit the information in the homologous sequences: i) a new generalised version of MERGE capable of employing any regressor as a base estimator; ii) the Tri-Training Regressor method, an adaptation of the Tri-Training method for regression problems. We find that the information inherent in the homologous sequences has the ability to improve the predictive capacity of models when the number of available sequences is scarce, especially when using the DCA encoding together with MERGE and an SVM regressor.","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"3 1","pages":""},"PeriodicalIF":8.6,"publicationDate":"2025-05-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144188911","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：