Journal of Cheminformatics最新文献

{"title":"Accurate prediction of activity cliff compounds based on bioactivity profiles depends on assay nearest neighbor relationships","authors":"Ryuto Abe,&nbsp;Tomoyuki Miyao,&nbsp;Jürgen Bajorath","doi":"10.1186/s13321-026-01210-9","DOIUrl":"10.1186/s13321-026-01210-9","url":null,"abstract":"<div><p>The definition of activity cliffs (ACs) depends on compound similarity and activity difference criteria and on activity data types. ACs are usually defined as pairs or groups of structurally similar compounds or structural analogues that are active against the same target, but have large differences in potency (requiring numerical potency values, preferably equilibrium constants). In addition, ACs have also been defined as pairs of structural analogues that are active or inactive in screening assays. In medicinal chemistry, ACs are of particular interest because they often reveal structure–activity relationship (SAR) determinants during compound optimization. In cheminformatics, ACs present challenging test cases for machine learning (ML) and activity predictions because they represent an extreme form of SAR discontinuity in compound data sets. Given their composition, ACs are notoriously difficult to predict based on chemical structure representations. Various attempts have been made to predict compound pairs forming ACs or the activity of AC compounds via ML, often reporting high accuracy. However, in the absence of data leakage from training to test sets, AC prediction accuracy based on chemical structure is low or moderate at best. As an alternative to structural representations, biological/functional compound descriptors might be considered such as biological assay profiles, which have been investigated for other compound activity prediction. In this work, we report the prediction of AC compounds based on bioactivity profiles derived from a compound profiling matrix using data partitioning schemes designed to control information or data leakage during ML. Under these stringent conditions, AC compound predictions based on bioactivity profiles often failed. However, we also observed subsets of highly accurate predictions and explored in detail why these predictions succeeded, but others failed. The analysis revealed a critically important role of assay similarity for successful AC compound predictions. Most profile assays did not measurably influence the predictions. By contrast, accurate predictions mostly depended on the presence of one or at most a few profile assays that were similar to test assays. In most cases, these profile assays could be identified and exploited for predictions by nearest neighbor searching, thus putting ML performance into perspective.</p><p><b>Scientific contribution</b> </p><p> As an alternative to the generally difficult prediction of ACs based on chemical structure, we introduce prediction of AC compounds based on bioactivity profiles. We show that accurate activity predictions of AC compounds do not depend on the global information content of assay profiles, but are largely determined by nearest neighbor relationships between profile and test assays.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":5.7,"publicationDate":"2026-06-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01210-9.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148154725","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Integrating multimodal features with deep learning for protein solubility prediction","authors":"Zechen Wang,&nbsp;Lai Heng Tan,&nbsp;Liangzhen Zheng,&nbsp;Jagath C. Rajapakse,&nbsp;Yuguang Mu","doi":"10.1186/s13321-026-01225-2","DOIUrl":"10.1186/s13321-026-01225-2","url":null,"abstract":"<div><p>Protein solubility prediction holds significant importance in the fields of biotechnology and medicine. With the continual advancements of computational and experimental techniques such as protein design, enzyme mining, and directed evolution, accurate prediction of native and mutant protein solubility has become a key step in accelerating the development of functional proteins. In this study, we extracted physicochemical properties and co-evolutionary features from protein sequences, and further incorporated graph-based protein representations along with surface features as inputs for solubility prediction. Building upon these features, we developed two models, ProSolNet and ProSolNet_Mut. ProSolNet predicts whether a protein is soluble, while ProSolNet_Mut predicts solubility changes induced by mutations. Compared with state-of-the-art models, both models achieved higher accuracy in their respective tasks. In addition, we investigated the underlying mechanisms and application potential of the models through interpretability analysis. </p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-06-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13449401/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148154738","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MolMeDB RDF: transforming a relational database about biological membranes to the RDF format to increase interoperability","authors":"Dominik Martinát,&nbsp;Jakub Juračka,&nbsp;Jakub Galgonek,&nbsp;Karel Berka","doi":"10.1186/s13321-026-01208-3","DOIUrl":"10.1186/s13321-026-01208-3","url":null,"abstract":"<div><p>Data on molecular behaviour within biological membrane models are pivotal for predictive pharmacology, particularly for modelling bioavailability and target engagement. MolMeDB provides a curated collection of such compound-membrane interaction data. To facilitate automated reasoning and data federation, we present a semantic transformation of MolMeDB into Resource Description Framework (RDF) format, exposed through a public SPARQL endpoint. The resulting schema incorporates community-standard ontologies—including those utilised by PubChem and ChEMBL—to maximise semantic interoperability and enable complex, cross-database federated queries.</p><p><b>Scientific contribution</b></p><p>MolMeDB RDF provides data on interactions between molecules and membranes and molecules and transporters in machine readable graph format with links to other RDF databases. SPARQL endpoint servicing this dataset enables flexible federated querying including structural queries utilizing SACHEM.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-06-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13449323/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148147936","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Chemical space visualization at scale: a survey of end-to-end pipelines and dataset-size archetypes","authors":"Maha M. AlShammari,&nbsp;Moayad Alnammi,&nbsp;Moataz Ahmed","doi":"10.1186/s13321-026-01223-4","DOIUrl":"10.1186/s13321-026-01223-4","url":null,"abstract":"<div><p>Chemical space visualization supports exploration of high-dimensional molecular data by revealing patterns of similarity, diversity, and structure–property relationships. As chemical libraries expand from thousands to billions of compounds, practical visualization increasingly depends on pipelines that balance chemical meaning with computational and memory constraints. In this survey, we review 56 studies published between 2000 and 2025 and synthesize the end-to-end workflow of chemical space visualization across five core stages: dataset choice, molecular featurization, dimensionality reduction (DR), clustering, and evaluation. We quantify usage trends over time and relate method selection to dataset scale. Across the literature, fingerprints remain the dominant representation for large libraries due to their scalability, while recent studies increasingly incorporate fragments, SMILES-based encodings, and learned embeddings when richer signals are needed. DR practice shows a shift from PCA-centric baselines to neighborhood-preserving methods such as t-SNE and UMAP, with graph layout approaches like TMAP enabling visualization at extreme scale. Clustering shows the weakest convergence to a single standard: hierarchical methods, K-means, and SOM remain frequent choices, complemented by scalable summarization and domain-driven strategies (e.g., BIRCH/BitBIRCH and scaffold-based partitioning) when all-pairs similarity becomes prohibitive. Finally, we propose dataset-size-aware pipeline archetypes and identify open challenges, including inconsistent structure-aware evaluation, limited reproducibility reporting, and the need for scalable, chemically grounded methods for ultra-large libraries.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-06-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13450572/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148147919","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Assessing the factors influencing the quality of pocket-conditioned 3D generative models","authors":"Kunyu Wang,&nbsp;Helen Lai,&nbsp;Ross Irwin,&nbsp;Jon Paul Janet,&nbsp;Alessandro Tibo","doi":"10.1186/s13321-026-01230-5","DOIUrl":"10.1186/s13321-026-01230-5","url":null,"abstract":"<div><p>Structure-based drug discovery (SBDD) aims to identify novel molecules that bind to therapeutic protein targets. The vast chemical space and limitations of traditional approaches make this task challenging. Recent advances in AI-generative models, such as flow matching, can produce novel, pocket-conditioned molecular structures directly in three-dimensional space. However, most pocket conditioned models in the literature are trained on structures derived from the Protein Data Bank (PDB), which contains structures with varying quality and inconsistent annotation. Moreover, the PDB is enriched with cofactors and natural products, thereby poorly representing real world SBDD scenarios. The relatively limited number of ligand series within the same pockets also hinder the model’s ability to learn protein-ligand interactions effectively. Here for the first time we report the results of training pocket-conditioned generative models on internal crystallography data from a large pharmaceutical company. We also investigate other key determinants of model performance, such as inclusion of hydrogens and pretraining on unconditional data. We evaluate how each factor affects the generative quality of the ligands across the diverse training settings. Our results provide practical guidelines for the development of more effective 3D generative models for SBDD and highlight key directions for future research toward reliable, pocket-aware molecular design.  </p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01230-5.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148100751","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards the design of artificial sensing materials via quantum-informed explainable AI","authors":"Li Chen,&nbsp;Leonardo Medrano Sandonas,&nbsp;Shirong Huang,&nbsp;Alexander Croy,&nbsp;Gianaurelio Cuniberti","doi":"10.1186/s13321-026-01232-3","DOIUrl":"10.1186/s13321-026-01232-3","url":null,"abstract":"<div><p>Computational design of sensing materials remains fundamentally challenging due to the vast configurational landscape and the absence of robust property correlations that enable efficient molecular representation. These challenges become particularly critical in healthcare applications, where reliable and interpretable molecular recognition is essential for non-invasive diagnostics. To address these challenges, we developed MORE-ML, a quantum-informed AI framework that combines electronic-structure-derived properties of e-nose molecular building blocks with machine learning (ML) methods to uncover sensing mechanisms and guide the design of new systems. Within this framework, we expanded our previous dataset, MORE-Q, to MORE-QX by sampling a larger conformational space of interactions between body odor volatilomes (BOV) molecules and mucin-derived receptors, both in the gas phase and when deposited on graphene. MORE-QX provides extensive electronic binding features (BFs) computed upon BOV adsorption. Analysis of the property space revealed weak correlations between quantum-mechanical (QM) properties of building blocks and resulting BFs. Leveraging this observation, we defined electronic descriptors of building blocks as inputs for tree-based ML models to predict BFs. Benchmarking showed CatBoost models outperform alternatives, especially in transferability to unseen compounds. Through explainable AI, we reduced the high-dimensional QM property space to a compact and physically interpretable set of descriptors, revealing the properties that most influence BF predictions. Collectively, MORE-ML combines QM insights with ML to provide mechanistic understanding and rational design principles for artificial sensing materials in BOV sensing. This approach establishes a foundation for advancing materials capable of analyzing complex odor mixtures, bridging the gap between molecular-level computations and practical e-nose applications.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":5.7,"publicationDate":"2026-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01232-3.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148051879","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Helmkit: fast and robust conversion of HELM notation to atomistic representations for large-scale macromolecular informatics","authors":"Ramon Adàlia,&nbsp;Gemma Sanjuan,&nbsp;Tomàs Margalef,&nbsp;Ismael Zamora","doi":"10.1186/s13321-026-01218-1","DOIUrl":"10.1186/s13321-026-01218-1","url":null,"abstract":"<div><p>The Hierarchical Editing Language for Macromolecules (HELM) provides a powerful framework for representing complex biomolecules, including peptides, oligonucleotides, and hybrid constructs, but existing tools for converting HELM notations to atomistic models suffer from limitations in speed, scope, and robustness. We introduce helmkit, an open-source Python library that enables direct, high-throughput conversion of HELM strings to RDKit molecular objects. Designed for general macromolecular structures, helmkit supports peptides, nucleic acids, chemical linkers, and hybrids, while natively handling inline monomers, special characters in names, and automatic inference of missing attachment points. Its streamlined architecture, with minimal dependencies and built-in parallelization, achieves processing speeds of up to 5,000 HELM entities per second. Validation on large-scale datasets from PubChem (878,442 entries) and CycPeptMPDB (7,298 entries) demonstrates near-perfect accuracy, with helmkit successfully parsing structures that fail in other libraries. By facilitating efficient, scalable analysis of diverse macromolecules, helmkit advances computational workflows in drug discovery, virtual screening, and biomolecular engineering. </p><p><b>Scientific contribution</b></p><p>We introduce a lightweight Python library that provides direct and accurate conversion from HELM notation to atomistic RDKit representations, addressing a practical limitation of existing cheminformatics tools. Unlike prior approaches, the library supports peptides, RNA, chemical linkers, and hybrid molecules within a single framework, while remaining robust to real-world HELM variants through built-in error correction and R-group inference. Validation on public datasets and high-throughput performance demonstrate that the method is reliable and suitable for large-scale molecular data processing.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13449311/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148051851","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Efficient fine-tuning of vision-language adapters in chemical VLMs for molecular image-text tasks","authors":"Hyukjun Choi,&nbsp;Donghyeon Lee,&nbsp;Juyoung Kang","doi":"10.1186/s13321-026-01214-5","DOIUrl":"10.1186/s13321-026-01214-5","url":null,"abstract":"<div><p>Vision–language models (VLMs) have recently shown strong potential for multimodal scientific applications, yet their direct application to molecular structure images remains challenging due to the need for precise visual–textual alignment. In this study, we investigate efficient fine-tuning strategies for adapting a large-scale VLM to molecular image captioning tasks. Our approach is built upon the Qwen3.5—9B architecture and focuses on lightweight parameter-efficient adaptation.</p><p>Specifically, we apply Low-Rank Adaptation (LoRA) to the language-side MLP layers and explore selective tuning of the visual merger module that bridges the vision encoder and the language model. This module compresses visual features and projects them into the language embedding space, making it a critical component for cross-modal alignment. Experimental results on molecular image captioning tasks show that the combined L-MLP + Visual Merger tuning strategy achieves the best performance while maintaining low computational overhead.</p><p>In addition, we evaluate multiple decoding strategies, including greedy decoding and stochastic sampling methods, to identify the most suitable inference configuration for molecular image captioning. The results demonstrate that selective fine-tuning of the visual–language interface significantly improves caption accuracy while preserving efficiency. These findings highlight the importance of adapting cross-modal projection modules when deploying general-purpose VLMs in specialized scientific domains such as cheminformatics.</p><p><b>Scientific contribution</b></p><p>We provide a systematic analysis of layer-wise fine-tuning strategies in large-scale vision–language models(VLMs), identifying which components are most critical for adapting general-purpose models to molecular image captioning tasks.</p><p>We identify the cross-modal projection module (visual merger) as a key factor in visual–textual alignment, demonstrating that its selective adaptation plays a central role in improving caption quality.</p><p>We propose a parameter-efficient fine-tuning strategy that combines light weight language-side adaptation with targeted tuning of the visual–language interface, achieving an effective balance between performance and computational cost.</p><p>We provide practical insights into efficient adaptation of VLMs in scientific domains, highlighting key considerations for achieving performance under computational constraints.</p></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13449681/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148051849","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Integrating chemical structure and high-throughput transcriptomics for mechanistically interpretable Tox21 bioactivity prediction","authors":"Guillaume Cattebeke,&nbsp;Anne-Sofie Vermeersch,&nbsp;Davie Cappoen,&nbsp;Dieter Deforce,&nbsp;Filip Van Nieuwerburgh","doi":"10.1186/s13321-026-01231-4","DOIUrl":"10.1186/s13321-026-01231-4","url":null,"abstract":"<div><p>Predictive toxicology increasingly emphasizes methods that combine scalable chemical screening with biologically interpretable mechanistic information. Existing computational approaches, however, rely largely on chemical structure alone and often fail to capture the cellular programs underlying compound-induced cellular responses. Herein, we describe a multimodal modeling framework that integrates chemical fingerprints with high-throughput transcriptomic (HTTr) dose–response profiles to predict activity for 41 curated Tox21 assay endpoints. HTTr data were obtained from TempO-Seq screens in MCF-7, U-2 OS, and HepaRG cells following exposure to ToxCast compounds across an eight-point concentration series ranging from 0.03 to 100 µM. Using gradient-boosted decision trees and nested compound-aware cross-validation, 13 assays achieved robust performance (mean area under the precision–recall curve (AUPRC) &gt; 0.75), spanning nuclear receptor signaling, stress-response pathways, and xenobiotic metabolism. SHapley Additive exPlanations (SHAP)-based feature attribution analysis showed that predictions depend on both structural motifs and transcriptional programs, in a manner consistent with established mechanistic relationships between chemical structure, nuclear receptor biology, and adaptive cellular responses. These findings illustrate how structure and high-throughput transcriptomic dose–response signature integration enables models that are accurate and mechanistically grounded, shifting computational toxicology toward transparent and biologically informed mechanistic bioactivity prediction.</p><h3>Graphical abstract</h3>\u0000<div><figure><div><div><picture><source><img></source></picture></div></div></figure></div></div>","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1186/s13321-026-01231-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148013043","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Predicting fire consequences with the transformer model based on multimodal feature fusion","authors":"Yilin Wang,&nbsp;Beibei Wang,&nbsp;Xiaolu Wang,&nbsp;Yichen Zhang,&nbsp;Jiquan Zhang,&nbsp;Yijie Song","doi":"10.1186/s13321-026-01228-z","DOIUrl":"10.1186/s13321-026-01228-z","url":null,"abstract":"&lt;div&gt;&lt;p&gt;Accurate prediction of fire consequences is fundamental to process safety management and quantitative risk assessment in the chemical process industries. Traditional empirical and computational fluid dynamics models often struggle to balance computational efficiency with predictive accuracy, especially when considering the complex interplay between molecular properties and process conditions. To address this challenge, this study proposes a high-precision Transformer-based prediction model for estimating the radiation effect distance, a critical parameter for defining safety zones. The model is founded on a multimodal feature fusion approach, integrating SMILES encoding, molecular descriptors, and three-dimensional molecular structures to comprehensively represent chemical identity and hazard potential. A dataset was generated using industry-standard PHAST software simulations, covering diverse release scenarios for 40 flammable chemicals. The proposed multimodal Transformer model significantly outperforms traditional machine learning models. Compared with the random forest (RFR) benchmark model, its coefficient of determination (R&lt;sup&gt;2&lt;/sup&gt;) has increased by approximately 15%, rising from 0.84 to 0.98. More importantly, it performs better in regression performance. The mean squared error (MSE) on the test set is as low as 0.0082, the root mean squared error (RMSE) is 0.0959, and the mean absolute error (MAE) is 0.0047. SHAP analysis confirmed the model's decision-making aligns with safety principles, identifying leak size and pressure as key hazard drivers. This confirms the model's reliability. This work successfully applies the Transformer architecture to fire consequence modeling within a process safety framework, achieving an optimal balance between computational performance and prediction accuracy. It provides a reliable and interpretable tool for enhancing industrial hazard assessment, supporting safety design, and informing emergency planning, thereby underscoring the transformative potential of deep learning in advancing chemical safety practices.&lt;/p&gt;&lt;p&gt;&lt;b&gt;Scientific contribution&lt;/b&gt;&lt;/p&gt;&lt;p&gt;This study proposes a Transformer-based deep learning framework with multimodal feature fusion for predicting firethermal radiation distances, establishing an innovative bridge between molecular informatics and process safety engineering. The model integrates SMILES sequences, three-dimensional molecular structures, quantum-chemical descriptors, and key process parameters into an end-to-end architecture, moving beyond traditional QSPR/QPCR methods that rely solely on handcrafted descriptors. Systematic comparison demonstrates that incorporating 3Dmolecular spatial information significantly improves prediction accuracy, achieving an 18.3% reduction in meanabsolute error for late pool fires. The proposed model substantially outperforms conventional machine learning benchmarks, attaining an R² of 0.98 for jet fires—a 15% improvement over random ","PeriodicalId":617,"journal":{"name":"Journal of Cheminformatics","volume":"18 1","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13397775/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148012997","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}