Journal of Chemical Information and Modeling 最新文献

{"title":"Interpretive Limits of Cross-Split AUPRC Differences in Deep Learning-Based Antibiotic Screening","authors":"Zhijie Pan,&nbsp;Tianming Han,&nbsp;Wenchi Ge,&nbsp;Churong Wang,&nbsp;Qi Zhao","doi":"10.1021/acs.jcim.6c03208","DOIUrl":"10.1021/acs.jcim.6c03208","url":null,"abstract":"<p>Deeplearning expands the chemical space accessible to antibioticdiscovery, making interpretation of evaluation results increasinglyimportant when only a few compounds can be prioritized from vast candidatelibraries. AUPRC depends on test-set prevalence, while molecular splittingcan alter training-to-test structural relatedness and test-set classcomposition, so when prevalence differs across splits, a cross-splitAUPRC difference alone cannot quantify how much of the performancechange is attributable to structural differences. Cross-split comparisonsshould therefore report test-set size, positive count, prevalence,splitting method, and the precision-recall summary metric, while directmodel comparisons should use the same fixed independent test set wheneverpossible. Claims about practical screening utility should also considertop-k performance, early enrichment metrics, or prospective experimentalhit rates under a prespecified budget, so that evidence remains alignedwith the scientific claims it is intended to support.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13703–13707"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149210138","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":"Synergy-Guided Molecular Redesign of Predicted P-Glycoprotein Substrates toward Synthetically Feasible Non-Substrates","authors":"Soyeon Jeon,&nbsp;Sung Wook Moon,&nbsp;Hyun Woo Kim","doi":"10.1021/acs.jcim.6c01938","DOIUrl":"10.1021/acs.jcim.6c01938","url":null,"abstract":"<p>Drug efflux mediated by P-glycoprotein (P-gp) remainsa major barrierto effective small-molecule therapeutics, limiting both bioavailabilityand central nervous system exposure. While machine learning modelscan predict P-gp substrates, translating these predictions into actionablemolecular design remains largely unexplored, particularly in enablingthe rational conversion of substrates into nonsubstrates to mitigateefflux liability. Here, we introduce the synergy-guided fragment geneticalgorithm (SF-GA), a computational framework that transforms predictiveinsight into an interpretable molecular design strategy. By integratingfragment synergy analysis into the mutation stage of a genetic algorithm,SF-GA biases fragment replacement toward co-occurring substructuresassociated with nonsubstrate classification, enabling interpretabilityat the chemical structure level and property-directed molecular optimization.Compared with genetic-algorithm and reinforcement-learning baselinesunder a fixed function-evaluation budget, SF-GA achieves competitiveoptimization performance while producing molecules with the highestretrosynthetic accessibility among all methods tested. Unlike thebaselines, every accepted mutation is reported together with the discardedand inserted fragments and their synergy scores, providing a fragment-levelrationale for each structural change; ablation against an unguidedand an importance-only control isolates the contribution of bit co-occurrence.The method enables scaffold-preserving and similarity-constrainedediting that converts P-gp substrates into predicted nonsubstrates.Docking against two human P-gp structures reveals weaker predictedbinding for molecules with higher predicted nonsubstrate probability,motivating future experimental evaluation of P-gp transport.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14174–14187"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149218796","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":"GLH2Syn: Global–Local Dual-View Heterogeneous Hypergraph Learning for Drug Synergy Prediction","authors":"Wen Zhang,&nbsp;Yuxuan Tian,&nbsp;Hang Qiu","doi":"10.1021/acs.jcim.6c02413","DOIUrl":"10.1021/acs.jcim.6c02413","url":null,"abstract":"<p>Drugcombination therapy has emerged as a promising strategy forimproving therapeutic efficacy. While hypergraphs have demonstratedsubstantial potential for high-order drug synergy modeling, existingapproaches still suffer from two critical limitations: (1) relianceon locally determined aggregation limits the utilization of multihoptopological context and broader structural information across thehypergraph; (2) homogeneous message passing overlooks the structuralheterogeneity of drugs and cell lines within hyperedges. To addressthese issues, we propose GLH²Syn, a global–localdual-view heterogeneous hypergraph learning framework for drug synergyprediction. Specifically, GLH²Syn operates on a heterogeneoushypergraph in which each drug–drug–cell line tripletforms a hyperedge, incorporating two complementary views: a globaldiffusion view and a local topology view. The global diffusion viewperforms diffusion over the node–hyperedge topology to deriveglobally informed incidence weights, thereby incorporating structuralcontext into local aggregation. The local topology view captures fine-grainedstructural heterogeneity within hyperedges through structure-awarenode-hyperedge propagation, distinguishing the heterogeneous rolesof drugs and cell lines in high-order interactions. Furthermore, wedesign a joint optimization objective integrating drug synergy predictionwith similarity reconstruction to preserve both predictive accuracyand biological consistency. Extensive experiments on two public datasets demonstrate that GLH²Syn consistently outperforms11 competitive baselines on both classification and regression tasksacross three test scenarios. Case studies further validate its potentialin discovering novel synergistic drug combinations.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13887–13901"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149221628","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":"Small-Molecule 1H and 13C NMR: From Chemical-Shift Prediction to Automated Structure Elucidation","authors":"Tingcai Dong,&nbsp;Chuanling Deng,&nbsp;Liqiu Tai,&nbsp;Meng Zhang,&nbsp;Yongfan Zi,&nbsp;Li Zhu,&nbsp;Fengmei Zhang,&nbsp;Le Cai","doi":"10.1021/acs.jcim.6c02915","DOIUrl":"10.1021/acs.jcim.6c02915","url":null,"abstract":"<p>¹H and ¹³C NMR chemical-shift predictionhas progressed from empirical rules and experimental databases tograph neural networks, three-dimensional molecular representations,and approaches that integrate quantum chemistry with machine learning;its applications have likewise expanded from chemical-shift calculationfor known structures to experimental spectral assignment, candidate-structurediscrimination, and automated structure elucidation. Forward structure-to-NMRprediction can generate predicted spectral-library data for candidatemolecules and support structure retrieval and candidate screeningby matching predicted and experimental chemical shifts; inverse NMR-to-structureanalysis progressively constrains the candidate space using structuralconstraints encoded in experimental spectra. This Review systematicallysurveys the development and growing integration of these two directions,focusing on how chemical-shift prediction contributes to peak-to-atomassignment, candidate-structure comparison, and automated structureelucidation, and examining how prediction error, candidate coverage,and experimental data quality affect final structural decisions. Morerecently, agentic approaches have begun to organize existing NMR predictionand structure-elucidation tools into iterative analytical systemsthat continuously update candidate structures from current spectroscopicevidence and select subsequent computational or experimental operations;further progress in automated structure elucidation depends on reliablepredicted spectral libraries, clearly defined model applicabilitydomains, and structural conclusions that can be tested against experimentalspectroscopic evidence.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13727–13752"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149248701","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":"QRxVision: Multimodal QR Encoding and Computer Vision for Domain-Specific Drug Discovery","authors":"Mithony Keng,&nbsp;Kenneth M. Merz, Jr.","doi":"10.1021/acs.jcim.6c01321","DOIUrl":"10.1021/acs.jcim.6c01321","url":null,"abstract":"<p>Computational approaches applied to the drug discoverypipelineare invaluable because of their overall speed or time to solution,sustainability, accessibility combined with ever improving accuracy.Furthermore, the high-throughput virtual screening supports practicalsampling of the vast chemical space for target-relevant bioactivecompounds that can be experimentally validated further downstreamin the drug development process. The current paradigm being practicedin computer-aided drug screening strategies centers around structure-basedand ligand-based approaches supported by a range of computationaltools including de novo design, molecular dynamics-based approaches,and machine learning (ML). Overall, the latter approach has showngreat promise generating models of high computational speed coupledwith good accuracy. In this work, we developed QRxVision, a domain-specificmultimodal ML model explicitly optimized for kinase target assignmentusing similarity-based profile matching and retrieval. QRxVision utilizesimage vision from a pretrained VGG16 (Visual Geometry Group–16)model for deep spatial feature extraction and cosine similarity calculation.By encoding drug multimodality-associated parameters (MAP) into standardizedQR code images, QRxVision can evaluate the similarity between testand reference compounds. Drug targets can then be assigned via transitiveprofile matching. Validation using kinase inhibitors shows that QRxVisionis 71% successful at assigning viable small molecules within the Top-5predictions (<em>p</em>-value &lt; 0.05). Overall, we believethat with the appropriate algorithmic modifications and domain retargeting,the QRxVision framework can be applied to other target assignmenttasks.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"13783–13794"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149275077","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":"Efficient and Accurate Ligand Strain Calculations in Solution with the AIMNet2 Neural Network Potential","authors":"Hatice Gokcan,&nbsp;Olexandr Isayev","doi":"10.1021/acs.jcim.6c02468","DOIUrl":"10.1021/acs.jcim.6c02468","url":null,"abstract":"<p>Conformational strain, the energy a ligand pays to adoptits boundconformation, contributes to binding affinity but is routinely misestimated.The standard estimate references a single gas-phase energy minimum.This ignores solvent and mishandles charged ligands: in vacuo theirintramolecular electrostatics collapse them into compact minima thatdeflate the unbound reference and inflate the apparent strain. Weintroduce AIMNet2-CPCM, a neural network potential trained to B97-3cenergies and forces in a conductor-like continuum solvent (CPCM).It treats neutral and charged species on the same footing, and itsbatched GPU inference makes a solvated global-minimum search practicalacross an entire screening library. Across ∼8K PDB ligandsthe solvated strain has a median of 4.0 kcal/mol and rises with netcharge. The trend is well determined through a net charge of −2.The −3 and −4 polyanions dominate the high-strain tail,where continuum solvation is least reliable, and their values arereported as upper bounds. Gas-phase force fields reproduce the neutralstrains but overstate the charged ones, because they collapse theunbound minimum that solvent screening keeps open. Adding a continuumsolvent to the force field removes most of this error and confirmsthe mechanism. Profiling at this scale further shows that the sameligand strains very differently across the targets that bind it: ligandstrain is a property of the bound pose, not a transferable moleculardescriptor.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14012–14024"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149226504","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":"AI-Driven Discovery Pipeline Identifies Carvedilol as a Novel Partial Agonist of THRβ","authors":"Shuai Han,&nbsp;Chaochun Wei,&nbsp;Han Ju,&nbsp;Tongyu Huo,&nbsp;Tzumei Wang,&nbsp;Yameng Liu,&nbsp;Chao Wang,&nbsp;Zhenming Liu,&nbsp;Xiaodong Dou,&nbsp;Ning Jiao","doi":"10.1021/acs.jcim.6c02366","DOIUrl":"10.1021/acs.jcim.6c02366","url":null,"abstract":"<p>Thyroid hormone receptor beta (THRβ) is a key therapeutictarget for metabolic dysfunction-associated steatotic liver diseaseand dyslipidemia, yet the discovery of structurally diverse agonistsremains challenging. Here, we developed an AI-driven hierarchicaldrug discovery framework integrating graph neural network (GNN)-basedvirtual screening, structure-based modeling, and multilevel biologicalvalidation to accelerate THRβ agonist identification. The optimizedGNN model outperformed conventional machine-learning approaches andenabled efficient prioritization of active compounds from an in-houselibrary. Subsequent molecular docking and MM/GBSA rescoring identified <strong>Carvedilol</strong> as a high-confidence candidate. TR-FRET assaysconfirmed that <strong>Carvedilol</strong> acts as a partial THRβagonist with preferential SRC2-2 coactivator recruitment, comparableto <strong>Resmetirom</strong>. Cellular assays further demonstrated that <strong>Carvedilol</strong> significantly reduced lipid accumulation and triglyceridelevels and upregulated key metabolic regulators, including <em>DIO1</em>, <em>ME1</em>, and <em>LDLR</em>. Moleculardynamics and dcTMD analyses elucidated that <strong>Carvedilol</strong> is stabilized within the ligand-binding pocket through dominanthydrophobic interactions and specific hydrogen bonds with Thr273 andHis435. Overall, this study repurposes <strong>Carvedilol</strong> asa structurally distinct THRβ agonist and validates an effectiveAI-guided strategy for nuclear receptor-targeted drug discovery.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14215–14230"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149241659","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":"Evaluating Drug–Drug Interactions by Constructing Advanced Machine Learning Models of ABC Transporter Inhibitors","authors":"Sean R. Carey,&nbsp;Xia Wen,&nbsp;Nada J. Daood,&nbsp;Elena Chung,&nbsp;Jacqueline B. Tiley,&nbsp;Lauren M. Aleksunes,&nbsp;Hao Zhu","doi":"10.1021/acs.jcim.6c02847","DOIUrl":"10.1021/acs.jcim.6c02847","url":null,"abstract":"<p>ATP-binding cassette(ABC) transporters play a critical role intransmembrane transport, and the inhibition of ABC transporter-mediatedefflux can lead to unintended drug–drug interactions (DDIs).The identification of new drug candidates as perpetrators of transporter-mediatedDDIs requires synthesis of significant drug products and costly experimentalevaluation. Alternatively, computational approaches can rapidly andaccurately screen the DDI liability of new drug candidates <em>in silico</em>. Here, we developed a novel DDI risk assessmentprotocol incorporating classic and graph-based quantitative structure–activityrelationship (QSAR) models. These models were trained on publiclyavailable, large-scale, structurally diverse <em>in vitro</em> chemical inhibition data for P-gp, BCRP, and MRP1. The resultantmodels yielded promising R² scores (0.58 to 0.82). Externalpredictions successfully identified inhibitors, verified by experimentalvalidation or literature review, and detected drugs with the potentialto perpetrate clinically relevant DDIs. In summary, we developed alarge, diverse database of ABC transporter inhibitors and presentan applicable alternative protocol that incorporates model predictionsto evaluate transporter-mediated DDIs for new drug candidates to preventpotentially harmful clinical outcomes.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14231–14246"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149251040","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":"An Integrated Computational-Experimental Workflow for Fungal Peroxygenase Discovery and Enzyme–Substrate Prediction","authors":"Victor J. Y. Lim,&nbsp;Yee-Song Law,&nbsp;Guangrong Peh,&nbsp;Clarence Tan,&nbsp;Charmaine Chia,&nbsp;Jhoann M. T. Miyajima,&nbsp;Nazreen Abdul Muthaliff,&nbsp;Pnelope Ng,&nbsp;Yoganathan Kanagasundaram,&nbsp;Wan Lin Yeo,&nbsp;Sebastian Maurer-Stroh,&nbsp;Yee Hwee Lim,&nbsp;Ee Lui Ang,&nbsp;Hao Fan","doi":"10.1021/acs.jcim.6c01050","DOIUrl":"10.1021/acs.jcim.6c01050","url":null,"abstract":"<p>Recent advances in enzyme informatics have produced increasinglypowerful general models for the prediction of protein function prediction.However, their performance remains limited for underexplored enzymefamilies where experimental data are scarce and heterologous expressionremains challenging. Here, we present a computational-experimentalworkflow for prioritizing, characterizing, and modeling fungal unspecificperoxygenases (UPOs) as a case study in enzyme discovery under data-limitedconditions. The workflow integrates large-scale sequence mining, expression-awarecandidate prioritization, systematic experimental screening, and machine-learning-basedprediction of enzyme–substrate activity. From more than 3000mined candidate sequences, 59 putative UPOs were shortlisted, expressedin <em>Escherichia coli</em> across two batches, and testedagainst colorimetric substrates, yielding 27 UPOs with consistentexpression and activity. A subset of 15 novel UPOs and 4 well-characterizedreference enzyme sequences was then screened against a chemicallydiverse panel of 27 substrates to generate an enzyme–substrateactivity data set. Using this data set, we developed two complementarymachine learning protocols based on gradient boosting: (i) a structure-informedmodel incorporating homology modeling, docking, and interaction-derivedfeatures (modeling-docking-ML) and (ii) a latent-representation modelutilizing joint protein/chemical language model embeddings (protein/chemical-language-model-ML).In retrospective validation, the models achieved performance witha Matthews correlation coefficient (MCC) of up to 55% and accuracyup to 78%. In selected prospective testing on previously unseen batch2 enzymes, the models achieved an MCC of up to 48% and accuracy upto 70%. Taken together, these results demonstrate that targeted experimentaldata generation, combined with complementary structure-informed andembedding-based models, can support prioritization and activity predictionwithin poorly characterized enzyme families while highlighting thecontinued challenge of generalizing to broader chemical substratespace.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14247–14258"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149210100","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":"Design Trade-Offs between Target Binding and Self-Association Revealed by De Novo SARS-CoV-2 Binders","authors":"Lourdes Carcelén,&nbsp;Daniel Peñas-Utrilla,&nbsp;Diego S. Ferrero,&nbsp;Cristina Ferrer-Orta,&nbsp;Marta Nadal,&nbsp;Francesco Colizzi,&nbsp;Juana C. Muñoz,&nbsp;Marta Taulés,&nbsp;Núria Verdaguer,&nbsp;Enrique Marcos","doi":"10.1021/acs.jcim.6c02394","DOIUrl":"10.1021/acs.jcim.6c02394","url":null,"abstract":"<p><em>De novo</em> protein binders have emergedas a powerfulplatform for molecular recognition, yet achieving high affinity whilemaintaining favorable biophysical properties remains a major challenge.Here, using the SARS-CoV-2 receptor-binding domain (RBD) as a modeltarget, we generated two <em>de novo</em> protein binder designcampaigns that deliberately imposed different trade-offs between bindingand monomeric behavior, and benchmarked them against a set of experimentallyvalidated binders. Through integrated experimental and computationalanalyses, we show that binder design is constrained by two distinctfailure modes that require different optimization strategies. Noneof the 22 original designs showed measurable binding to the RBD, withone class remaining monomeric but failing to achieve sufficient targetbinding, while the other had more competent binding interfaces butexhibited excessive surface hydrophobicity that promoted self-association.By integrating AlphaFold-based predictions, molecular dynamics simulations,docking analyses, and surface aggregation propensity (SAP) calculations,we establish complementary computational signatures that distinguishthese failure modes and reveal their underlying structural determinants.Guided by these insights, we prospectively redesigned a representativefailed binder by reducing noninterface hydrophobicity while preservingfolding and binding interface quality. Experimental characterizationby SEC-MALS, surface plasmon resonance, and circular dichroism confirmedthat the redesigned variants were stable monomers and acquired lowμM binding affinities. Together, our results define the mechanistictrade-offs between binding and self-association in <em>de novo</em> protein binder design and provide a framework for diagnosing andrationally overcoming common design failure modes.</p>","PeriodicalId":44,"journal":{"name":"Journal of Chemical Information and Modeling ","volume":"66 21","pages":"14053–14067"},"PeriodicalIF":6.4,"publicationDate":"2026-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149251041","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}