Journal of Chemical Information and Modeling 最新文献

{"title":"In Silico Design of Dimerization Interfaces in Drosophila melanogaster Glutathione Transferase","authors":"Nicolas Petiot,Mathieu Schwartz,Jeanne Chaloyard,Patrice Delarue,Patrick Senet,Jean-Marie Heydel,Fabrice Neiers,Adrien Nicolaï","doi":"10.1021/acs.jcim.6c01706","DOIUrl":"https://doi.org/10.1021/acs.jcim.6c01706","url":null,"abstract":"Understanding the molecular determinants of protein–protein interface stability remains a central challenge for the rational engineering of oligomeric enzymes. Here, the glutathione transferase GSTE3 from Drosophila melanogaster is used as a model system to decipher the factors governing dimer stability through an integrated structural, computational, and experimental approach. The X–ray crystal structure reveals a canonical homodimeric organization with a conserved Epsilon-class dimerization motif at the subunit interface. Molecular dynamics simulations combined with inter-residue contact network analysis provide a quantitative description of the interface, identifying persistent and transient interactions that contribute to dimer stability. Building on this framework, sequence-space exploration by Monte Carlo simulations is employed to design the dimerization interface and enhance stability. Distinct modeling strategies, differing in mutation schemes and energy evaluation methods, yield multiple classes of variants with diverse interfacial properties. Their stability and interaction patterns are further assessed using molecular dynamics simulations, enabling comparative analysis of structural robustness and contact network organization. Selected variants are subsequently produced in vitro to evaluate their experimental feasibility. Altogether, this study establishes a multiscale framework for the rational engineering of protein–protein interfaces and highlights both the potential and limitations of sequence–based optimization strategies for stabilizing enzyme dimers.","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"6 1","pages":""},"PeriodicalIF":5.6,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149368318","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}

{"title":"Structure Prototype Learning for Federated Molecular Property Prediction under Scaffold Heterogeneity","authors":"Tingkai Sun,&nbsp;Xiucai Ye,&nbsp;Tetsuya Sakurai","doi":"10.1021/acs.jcim.6c02297","DOIUrl":"10.1021/acs.jcim.6c02297","url":null,"abstract":"<p>Federated molecular property predictionenables collaborationacrossprivate chemical libraries, but parameter averaging does not explicitlyaccount for scaffold heterogeneity. FedSPL groups molecules usingfixed descriptors and exchanges embedding prototypes alongside modelparameters, enabling label-independent prototype matching while retainingsupervised property prediction. Across five classification benchmarksand three heterogeneity levels, matched three-seed evaluation usingmasked, held-out server-model ROC-AUC showed higher mean final AUCthan FedAvg in seven of 15 settings. Consistently positive differenceson SIDER and mixed outcomes elsewhere indicate data set-dependentbenefits, motivating validation-guided use of structural informationin federated prototype learning.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13870–13886"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149221629","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}

{"title":"Systematic Profiling of Encoder Architectures and Multimodal Embeddings in Protein Foundation Models Reveals Data Set-Driven Separability Regimes in Prediction of Allosteric and Orthosteric Binding Sites","authors":"Alina Bazarova,&nbsp;Gennady M. Verkhivker","doi":"10.1021/acs.jcim.6c03114","DOIUrl":"10.1021/acs.jcim.6c03114","url":null,"abstract":"&lt;p&gt;Allosteric regulation represents a fundamental mechanismof proteinfunction, yet distinguishing allosteric from orthosteric protein-bindingsites remains a persistent computational challenge. While multimodalprotein foundation models offer the potential to integrate complementarybiological signals, including sequence, structure, functional annotations,and conformational dynamics, their performance determinants in allostericbinding site detection remain poorly understood. We introduce a unifiedcomputational framework for profiling multimodal protein foundationmodels across distinct binding site separability regimes. Rather thanevaluating models solely by predictive accuracy, the framework combinessystematic modality embedding ablations, encoder architecture comparisons,and variance decomposition to characterize how evolutionary, structural,functional, and dynamical information contributes to allosteric sitediscrimination. Using the OneProt multimodal model, we evaluate twocomplementary levels of multimodal integration: (a) encoder architecturesthat differ in the modalities incorporated during pretraining and(b) downstream combinations of pocket, sequence, and text embeddingsused for classification. To systematically probe the determinantsof model performance, we benchmark these configurations across fourassembled data sets of protein complexes representing a spectrum ofbiological complexity and a range of structural, dynamic, and evolutionarycontexts for orthosteric and allosteric binding sites. Through comprehensiveembedding ablations, encoder architecture comparisons, and variancedecomposition, we demonstrate that model performance is governed primarilyby intrinsic data set properties rather than architectural complexity,with data set identity accounting for 63.7% of explainable variance.Across the data sets examined, we identify substantial differencesnot only in overall separability but also in the information accessiblefrom different pretrained representations. Single-modality controlsshow that high predictive performance does not necessarily imply relianceon the explicitly provided pocket representation: depending on dataset composition, substantial discrimination can already be obtainedfrom sequence- or text-based multimodal representations. Within-proteincomparisons nevertheless demonstrate that pocket-containing representationsretain binding site-specific discriminatory information when proteinidentity and the associated sequence and text representations areheld constant. At the encoder level, structural and molecular dynamicsencoders provide the greatest benefit in intermediate and high separabilitysettings. Example-level analysis further shows that combining datasets can reshape which binding sites are preferentially recognized.These findings establish that the success of multimodal foundationmodels depends critically on alignment between available modalitiesand the biological signatures that distinguish regulatory classesin each data ","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13913–13937"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149248652","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}

{"title":"Benchmarking and Experimental Validation of Machine Learning Strategies for Enzyme Engineering","authors":"Zishuo Zeng,&nbsp;Jiao Jin,&nbsp;Rufang Xu,&nbsp;Yuzhen Wu,&nbsp;Xiaozhou Luo","doi":"10.1021/acs.jcim.6c01200","DOIUrl":"10.1021/acs.jcim.6c01200","url":null,"abstract":"<p>Enzyme-directed evolution increasingly relies on computationaltools to prioritize mutations, yet their practical value is difficultto assess because kinetic data are often aggregated across heterogeneousassay conditions, inflating apparent generalization. Here we introduceEnzyArena, a curated benchmark in which kinetic parameters (<em>kcat</em>, <em>Km,</em> or <em>kcat</em>/<em>Km</em>) are grouped by experimental campaign and measured undermatched conditions, thereby minimizing confounders such as pH, temperature,reaction, and batch. Using EnzyArena, we systematically evaluate 20models spanning four widely used strategies for guiding enzyme evolution:protein–ligand binding affinity prediction, protein stabilityprediction, zero-shot fitness prediction and enzyme kinetic parametersprediction. Across subsets derived from public databases and 25 literature-derivedmutagenesis data sets, most models exhibit weak, fragile, or inconsistentcorrelation with measured functional outcomes. Kinetic-parameter predictorsperform strongly on database-derived subsets but lose their advantageon literature-derived mutagenesis data sets, whereas zero-shot predictorsshow more consistent generalization. A simple consensus of multiplezero-shot models further improves the precision of identifying beneficialmutants in retrospective analyses. We prospectively validated thesefindings in a wet-lab campaign (150 mutants) comparing random mutants,UniKP-prioritized mutants, and ESM-1v-prioritized mutants (representingsupervised kinetic parameters prediction and zero-shot fitness prediction,respectively), where ESM-1v achieved the highest utility and UniKPunderperformed the random baseline. Together, this study establishesrealistic baselines for computational mutant prioritization and indicatesthat zero-shot strategies provide the most consistent, although stillmodest, performance among the approaches evaluated.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14038–14052"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149275078","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}

{"title":"FlexENN: A Graph Neural Network for Binding Energy Prediction of Globular and Intrinsically Disordered Proteins","authors":"Maryum Irshad,&nbsp;Kassandra Ori-McKenney,&nbsp;Ruxandra I. Dima","doi":"10.1021/acs.jcim.6c01545","DOIUrl":"10.1021/acs.jcim.6c01545","url":null,"abstract":"<p>Intrinsically disordered proteins (IDPs) drive a largefractionof cellular signaling, transcription, and assembly through interfacesthat often lack a defined geometry, making the prediction of theirbinding energies challenging for methods developed primarily for foldedcomplexes. Here, we introduce FlexENN, an interface-focused message-passinggraph neural network framework comprising two independently trainedmodels: FlexENN-F for folded protein complexes and FlexENN-I for complexescontaining a disordered partner. The FlexENN architecture constructsa local interface graph in which each residue node integrates therelative geometry, flexibility-related structural descriptors, andsequence information, allowing the network to learn structure-affinityrelationships from a single resolved bound conformation. We also constructan IDP-containing dataset for training the FlexENN-I model. Benchmarkedacross folded complexes and IDP-containing systems, including tubulin–tubulininterfaces within microtubules (MTs) and MTs in complex with microtubule-associatedproteins (MAPs), with reference binding affinity values extractedfrom our experiments, the models reproduce physically reasonable energeticprofiles, including the stronger interdimer relative to the lateralMT interactions, while feature-ablation analyses reveal system-dependentcontributions of selected descriptors to the predictions. More broadly,this work delivers an accurate, structure-aware approach to predictingbinding free energies for complexes of folded protein and complexesthat contain ordered or partially ordered IDP regions for which asufficiently well-defined bound structure is available.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13795–13808"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149151217","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}

{"title":"Kinase Inhibitors can Change Protonation or Tautomeric State upon Binding","authors":"Gehan A. Ranepura,&nbsp;Saqibul I. Chowdhury,&nbsp;Elizabeth A. Rosenzweig,&nbsp;Ariën S. Rustenburg,&nbsp;Raquel López-Ríos de Castro,&nbsp;Junjun Mao,&nbsp;John D. Chodera,&nbsp;Sukrit Singh,&nbsp;M. R. Gunner","doi":"10.1021/acs.jcim.6c02792","DOIUrl":"10.1021/acs.jcim.6c02792","url":null,"abstract":"<p>The binding affinity of a ligand to a protein is influencedbythe protonation and tautomeric states of both partners. However, thisrelationship remains under-investigated due to the limited availabilityof computational tools capable of treating charge and tautomer statestogether in a scalable manner to study clinically relevant systems.Here, we use multi-conformation continuum electrostatics (MCCE) tocalculate the protonation and tautomer distributions of nine kinasedomains bound to 18 FDA-approved inhibitors across 37 cocrystal structures,while considering their Boltzmann-ensemble. The full pipeline wasrepeated in triplicate, so that every reported value carries a measuredrun-to-run uncertainty. Our simulations show that protein net chargeand proton distribution remain largely stable even upon binding chargedinhibitors. Our results find that individual inhibitor charges aredynamic, frequently increasing or decreasing upon binding a specificprotein target. We observe that kinase-inhibitor binding shifts therelative probabilities of low-energy states (Δ<em>G</em> &lt; 2.8 kcal/mol), but does not increase the probability of higher-energystates in the bound form. By considering all low-energy charge statesand tautomers, we identify when two solution-state tautomers of nearlyequal population resolve upon binding into a single state, creatingfree energy differences of up to 3.5 kcal/mol between tautomers, andin some complexes differences beyond the resolution of the sampling.In turn, we find that minority species can become the dominant componentin the bound state, emphasizing the necessity of considering ensemble-wideprotonation and tautomer states to accurately predict protein–ligandbinding energetics.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14159–14173"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149210137","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}

{"title":"Reducing Data Requirements for Sequence–Property Prediction in Copolymer Compatibilizers via Deep Neural Network Tuning","authors":"Nishat Nayla Labiba,&nbsp;Md Mushfiqul Islam,&nbsp;Lawrence O. Hall,&nbsp;David S. Simmons","doi":"10.1021/acs.jcim.6c00178","DOIUrl":"10.1021/acs.jcim.6c00178","url":null,"abstract":"<p>Synthetic sequence-controlled polymers promise to transformpolymerscience by combining the chemical versatility of synthetic polymerswith the precise sequence-mediated functionality of biological proteins.However, design of these materials has proven extraordinarily challenging,because they lack the massive data sets of closely related evolvedmolecules that accelerate design of proteins. Here, we report on anArtificial Intelligence strategy to reduce the amount of data necessaryto accelerate the design of these materials. We focus on data connectingthe repeat-unit-sequence of a model compatibilizer molecule to itsability to reduce the interfacial tension between distinct polymerdomains. The optimal sequence of these molecules, which are essentialfor applications such as mixed-waste polymer recycling, depends stronglyon variables such as concentration and chemical details of the polymer.We show that a deep neural network trained on low-fidelity data forsequence/interfacial tension relations at one set of conditions canbe rapidly tuned to make higher-fidelity predictions at a distinctset of conditions, requiring far less data than would ordinarily beneeded within the data sets we have tested. This priming using a pretrainedmodel and then tuning it on unseen data from a different data setapproach could allow a single low-fidelity parent data set to accelerateprediction and design in a larger set of related systems.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13753–13767"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149218871","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}

{"title":"DockBind: A Quality-Controlled Data Set of Modeled Protein–Ligand Complexes Associated with Affinity Data","authors":"Alberto Rovetta,&nbsp;Alberto Memè,&nbsp;Hanne S. Antila,&nbsp;Ruth Brenk","doi":"10.1021/acs.jcim.6c02093","DOIUrl":"10.1021/acs.jcim.6c02093","url":null,"abstract":"<p>Ligand–protein interaction prediction is essentialfor structure-baseddrug discovery, but further development of machine-learning scoringfunctions is limited by a lack of high-quality training data sets.Here, we present DockBind, a curated data set of modeled protein–ligandcomplexes paired with experimental binding affinity data, generatedthrough a quality-controlled docking workflow. This workflow was developedbased on an expanded docking benchmark, where we systematically evaluatedfree and template-based docking methods as well as ligand similarityand maximum common substructure (MCS)-based pose filtering to generateand identify the most realistic poses. The resulting workflow wasthen incorporated into a pipeline integrating structural and bindingaffinity data from Binding MOAD, the Protein Data Bank, ChEMBL, andBindingDB, which resulted in 18,332 complexes across 612 protein targets,with high structural reliability and chemical diversity─theDockBind data set─providing a valuable resource for trainingand benchmarking machine-learning scoring functions.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13938–13955"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149248672","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}

{"title":"Modulating Transthyretin Fibril Stability with D-Retro-Inverso Peptides","authors":"Lucy M. Coleman,&nbsp;Ulrich H. E. Hansmann","doi":"10.1021/acs.jcim.6c02525","DOIUrl":"10.1021/acs.jcim.6c02525","url":null,"abstract":"<p>A major cause ofheart failure in elderly patients isthe depositionof transthyretin (TTR) fibrils. Using molecular dynamics simulations,we explore how the stability of TTR fibrils can be modulated by D-retro-inverso(DRI) peptides, built from D-amino acids with the sequence of theparent peptide switched. We screened a library of peptide candidatesgenerated from aggregation-prone regions in the TTR sequence, includingpeptides from WALTZ-DB 2.0 database of amyloid peptides, and selectedthree peptides based on their potential to disrupt interactions ina structurally flexible region of the fibril. Peptides were dockedto the surface of the TTR fibril using HADDOCK 2.4, followed by all-atommolecular dynamics simulations to evaluate peptide–fibril interactionsand their effects on fibril structure. We finally describe a potentialmechanism by which one of these peptides, DRI-K6V, perturbs the structuralintegrity of TTR fibrils, and identify the segment K80-S100 as a potentialtarget for modulating TTR fibril stability. Our results may open theway to the design of peptide drugs targeting established TTR amyloidosis.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14115–14125"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149250884","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}

{"title":"DBNAMP: A Chemically Curated Database of Noncanonical Antimicrobial Peptides with Standardized Molecular Representations","authors":"Tian Liang,&nbsp;Bai Li,&nbsp;Xiaokun Hong,&nbsp;Lin Yang,&nbsp;Jinyu Li","doi":"10.1021/acs.jcim.6c02717","DOIUrl":"10.1021/acs.jcim.6c02717","url":null,"abstract":"<p>Antimicrobial peptides (AMPs) are promising anti-infectiveagents,yet conventional sequence-centered resources often inadequately representnoncanonical residues, stereochemical variations, terminal modifications,cyclization, lipidation, stapling, and peptidomimetic scaffolds, limitingchemically meaningful annotation, comparison, and structure–activityrelationship analysis. Here, we present DBNAMP, a manually curatedand chemically explicit database of AMP-like entities containing residue-levelnoncanonical features, molecular-level covalent modifications, orboth; it currently comprises 1,859 curated entries, including 1,421entries with computed representative three-dimensional conformersand 129 standardized controlled-vocabulary records for noncanonicalresidues and peptide modifications. Among 537 source-level ambiguity-auditrecords, 84 remained unresolved after review, of which 82 retainedat least one uppercase X character. Only one such record was includedin the final release. Each DBNAMP entry integrates sequence, chemicalannotation, antimicrobial activity, target organisms, physicochemicaldescriptors, molecular representations, structural information, andliterature or patent provenance. In addition to browsing, filtering,and data download, DBNAMP enables chemically informed explorationof related peptide entities through standardized molecular representationsand incorporates a lightweight SMILES-based prioritization modulefor preliminary ranking of candidate molecules. By bridging sequence-levelpeptide annotation and chemically explicit molecular representation,DBNAMP supports chemoinformatics analysis, structure–activityrelationship studies, virtual screening, and computer-aided designof chemically diverse AMP candidates and is freely accessible at https://dbnamp.cn/.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13974–13985"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149251104","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}