# Fileset

[AdvMater.25_37_e10239.pdf](https://mdr.nims.go.jp/filesets/2b6fe203-5ac0-4891-8305-7f59e5e9085d/download)

## Creator

Hayeon Bae, Hyunsub Ji, Konstantin Konstantinov, Ronald Sluyter, [Katsuhiko Ariga](https://orcid.org/0000-0002-2445-2955), Yong Ho Kim, Jung Ho Kim

## Rights

[Creative Commons BY-NC Attribution-NonCommercial 4.0 International](https://creativecommons.org/licenses/by-nc/4.0/)

## Other metadata

[Artificial Intelligence‐Driven Nanoarchitectonics for Smart Targeted Drug Delivery](https://mdr.nims.go.jp/datasets/e8190a99-4e8f-4352-a69b-28d183c36428)

## Fulltext

Artificial Intelligence‐Driven Nanoarchitectonics for Smart Targeted Drug DeliveryPERSPECTIVEwww.advmat.deArtificial Intelligence-Driven Nanoarchitectonics for SmartTargeted Drug DeliveryHayeon Bae, Hyunsub Ji, Konstantin Konstantinov, Ronald Sluyter, Katsuhiko Ariga,Yong Ho Kim,* and Jung Ho Kim*The development of data-driven and targeted drug delivery systems isessential for advancing precision therapeutics. Despite substantialprogress in nanocarrier development, conventional platforms continueto face major challenges in clinical translation due to biological complexity,off-target accumulation, and limited adaptability to dynamic physiologicalenvironments. The integration of nanoarchitectonics and artificial intelligence(AI) offers an advanced strategy for engineering delivery systems that arestructurally programmable, stimuli-responsive, and autonomously optimized.Nanoarchitectonics enables the construction of hierarchical nanostructureswith precise spatial and temporal control, while AI facilitates modeling,prediction, and iterative optimization throughout the development pipeline. Inthis perspective, an AI-driven nanoarchitectonics framework is introduced fortargeted drug delivery, structured around three key phases: 1) molecular targetidentification through bioinformatic profiling, 2) machine learning (ML)-guidedsurface engineering to enhance targeting specificity, and 3) in silico modelingof delivery dynamics and systemic distribution. Drawing on recent advancesand representative case studies, how AI tools are illustrated, from generativedesign algorithms to predictive pharmacokinetic models, are transforming thefield from empirical formulation toward mechanism-informed and AI-drivenintelligent design. By highlighting current limitations and outlining futuredirections for the integration of AI and nanoarchitectonics, are concludedwith a focus on enabling clinically translatable nanomedicine platforms.H. Bae, K. Konstantinov, J. H. KimInstitute for Superconducting & Electronic Materials (ISEM)Faculty of Engineering and Information SciencesUniversity of Wollongong Innovation CampusSquires Way, North Wollongong, NSW 2500, AustraliaE-mail: jhk@uow.edu.auH.Bae,H. Ji, Y.H. KimDepartment ofNanoScience andTechnologySKKUAdvanced Institute ofNanotechnology (SAINT)SungkyunkwanUniversity (SKKU)2066Seobu-ro, Jangan-gu, Suwon,Gyeonggi-do 16419, Republic of KoreaE-mail: yhkim94@skku.eduThe ORCID identification number(s) for the author(s) of this articlecan be found under https://doi.org/10.1002/adma.202510239© 2025 The Author(s). Advanced Materials published by Wiley-VCHGmbH. This is an open access article under the terms of the CreativeCommons Attribution-NonCommercial License, which permits use,distribution and reproduction in any medium, provided the original workis properly cited and is not used for commercial purposes.DOI: 10.1002/adma.2025102391. IntroductionThe development of advanced drug deliverysystems plays an increasingly critical rolein enhancing therapeutic specificity andreducing off-target effects.[1] Among vari-ous strategies, targeted drug delivery hasemerged as a particularly promising ap-proach, enabling the precise localization oftherapeutic agents to diseased tissues orspecific cell populations.[2] Reducing off-target activity enhances both therapeuticprecision and tolerability. Yet, conventionalnanocarriers frequently fail to achieve op-timal performance in physiologically rele-vant conditions.[3] Poor responsiveness tobiological cues, inefficient tissue penetra-tion, suboptimal biodistribution, and pre-mature clearance or degradation frequentlylead to reduced therapeutic efficacy andclinical failure.[4]To address these challenges, a shiftin design philosophy is required. Thisshift involves moving away from tradi-tional material-centric approaches and to-ward strategies that emphasize complex-ity, hierarchy, and adaptability.[5] Nanoarchi-tectonics has emerged as a powerful andR. SluyterSchool of ScienceFaculty of ScienceMedicine and Health and Molecular HorizonsFaculty of ScienceMedicine and HealthUniversity of WollongongWollongong, NSW 2522, AustraliaK. ArigaDepartment of Advanced Materials ScienceGraduate School of Frontier SciencesThe University of Tokyo5-1-5 Kashiwanoha, Kashiwa, Chiba 277–8561, JapanK. ArigaResearch Center for Materials NanoarchitectonicsNational Institute for Materials ScienceNamiki 1-1, Tsukuba 305-0044, JapanY. H. KimDepartment of Nano EngineeringSungkyunkwan University (SKKU)2066 Seobu-ro, Jangan-gu, Suwon, Gyeonggi-do 16419, Republic of KoreaAdv. Mater. 2025, 37, e10239 e10239 (1 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbHhttp://www.advmat.demailto:jhk@uow.edu.aumailto:yhkim94@skku.eduhttps://doi.org/10.1002/adma.202510239http://creativecommons.org/licenses/by-nc/4.0/http://creativecommons.org/licenses/by-nc/4.0/http://crossmark.crossref.org/dialog/?doi=10.1002%2Fadma.202510239&domain=pdf&date_stamp=2025-08-07www.advancedsciencenews.com www.advmat.deFigure 1. Schematic representation of the integration between artificial intelligence (AI) and nanoarchitectonics for intelligent drug delivery. Createdwith BioRender.com.integrative strategy to meet this need. Defined as the method-ology of architecting functional systems by arranging atoms,molecules, and nanoscale units into well-organized structures,nanoarchitectonics bridges the gap between molecular designand macroscopic function.[6] It is a multidisciplinary frame-work that integrates nanotechnology, supramolecular chemistry,molecular self-assembly, and bioinspired design to constructfunctional nanostructures with precisely controlled architec-tures and dynamic behaviors.[7] Its approach emphasizes theintegration of diverse nanoscale processes, including molecu-lar manipulation, self-organization, and alignment under ex-ternal fields, into multistep, often hierarchical constructionstrategies.[8] Nanoarchitectonics enables the development of in-telligent nanocarriers capable of responding to microenviron-mental stimuli, performing logic-gated release, or reorganizingtheir structure in situ.[9] These systems exhibit emergent func-tionalities beyond the reach of conventional fabrication, position-ing nanoarchitectonics as a versatile platform for programmabledrug delivery.[10]The integration of nanoarchitectonics with artificial intel-ligence (AI) technologies opens new opportunities in drugdelivery.[11] AI encompasses powerful computational tools foranalyzing complex, high-dimensional datasets, including multi-omics profiles, structural databases, imaging data, and biosen-sor outputs, and for extracting actionable insights to guide ratio-nal material design.[12] Machine learning (ML), a major subset ofAI, enables data-driven model development and predictive analy-sis without explicit programming. Deep learning (DL), a special-ized branch of ML rooted in artificial neural networks (ANNs),supports simulation of nanocarrier–biological interactions, iden-tification of optimal surface chemistries, and prediction of invivo biodistribution patterns from structural features.[13] Thesecapabilities enable hypothesis generation, predictive modelingof physicochemical and pharmacokinetic properties, and au-tonomous optimization of nanocarrier design.[14] This significanttransition from traditional trial-and-error methods to computa-tionally guided, mechanism-informed workflows hold promisesfor advancing precision and efficacy of targeted drug deliverysystems.[15]This perspective introduces a unified framework in which AI-integrated nanoarchitectonics drives the development of next-generation targeted drug delivery systems. As illustrated inFigure 1, the framework emphasizes how artificial intelligencecontributes at each step of the nanoarchitectonic design pro-cess, from target identification to delivery optimization. The pro-posed system incorporates bioinformatic profiling for database-guided target discovery, computational design of nanocarriers,AI-driven engineering for targeted delivery, and in silico simula-tion to optimize therapeutic efficacy. Each component leveragesmachine learning and data-drivenmodeling tools to enable adap-tive and feedback-informed design. The figure presents a mod-ular and sequential workflow in which AI helps refine surfacechemistry, predict biological responses, and customize deliverystrategies for enhanced precision. This integrated and recursivearchitecture supports real-time optimization and reflects biolog-ical design principles, offering a systems-level approach to moreAdv. Mater. 2025, 37, e10239 e10239 (2 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.depersonalized and efficient nanomedicine. In the following sec-tions, we elaborate on each element of this framework, reviewrepresentative case studies, and discuss key challenges and fu-ture opportunities in advancing AI-enabled, dynamically tunabledrug delivery platforms.2. AI-Driven Workflow for Targeted Drug DeliverySystemsThe integration of AI into targeted drug delivery represents aparadigm shift from conventional empirical approaches to anintelligent and closed-loop design framework.[16] This shift en-ables a dynamic workflow where each phase is informed by dataand continuously refined through iterative feedback. The archi-tecture mimics the adaptability of biological systems while accel-erating therapeutic innovation. In the context of this framework,“adaptive” refers to three interrelated dimensions. First, materialadaptability describes nanocarriers that respond to environmen-tal stimuli such as pH, redox gradients, or enzymatic activity.[17]Second, computational adaptability reflects the iterative learningprocess of AImodels that refine design predictions based on newdata.[18] Third, system-level adaptability involves real-timemodu-lation of therapeutic parameters using patient-specific physiolog-ical inputs, such as those collected from wearable biosensors.[19]These adaptive layers enable smart, personalized drug deliverystrategies that dynamically respond to complex and evolving bio-logical conditions.The workflow begins with the rational design of nanocarrierstailored to specific therapeutic goals. Formulation follows, involv-ing the synthesis of nanoparticles with controlled size, charge,loading capacity, and release kinetics. These nanocarriers un-dergo comprehensive characterization and biological verificationthrough in vitro or in vivo studies, generating multi-dimensionaldatasets on performancemetrics such as uptake, biodistribution,and efficacy.[20] Data from each stage is systematically curated andintegrated into a comprehensive dataset. This dataset then un-dergoes processing, including normalization, feature extraction,and data curation, to prepare it for ML. Once processed, the dataare used to train and validate predictive models that learn com-plex relationships between input design parameters and biolog-ical outcomes. These models can then generate predictions fornew nanocarrier candidates with desirable delivery profiles.Predicted candidates re-enter the experimental cycle, begin-ning again with design and formulation, followed by character-ization and verification. This iterative loop drives continuous im-provement of predictive models and fosters the stepwise evolu-tion of nanocarriers, linking computational design to biologicalfunction. It facilitates the adaptive delivery systems that evolvethrough feedback and intelligently respond to data-driven designadjustments. The loop functions as the engine of a broader work-flow that is modular and scalable to diverse material types, ther-apeutic payloads, and disease contexts.As shown in Figure 2, the workflow operates as a self-improving, data-driven pipeline that accelerates the convergenceof material design, functional evaluation, and therapeutic ap-plication. By harnessing the capacity of AI to analyze and inte-grate high-dimensional data, this workflow facilitates the rapididentification of high-performance, personalized drug deliverysolutions while reducing experimental burden. It establishesthe operational foundation nanoarchitectonic approach to pre-cision nanomedicine. The general framework introduced hereunderpins the practical strategies discussed in the following sec-tions. These include AI-guided material selection, model-drivennanocarrier design, and iterative experimental validation, all ofwhich work together to realize and refine the proposed designloop.3. AI-Guided Process for Nanocarrier andNanomaterial DesignThe rational design of nanocarriers for targeted drug deliveryrequires a deep understanding of material properties, physico-chemical interactions, and biological behavior. Traditional for-mulation development has been largely empirical, often involv-ing time-intensive experimentation with limited predictability.The integration of AI, particularly ML, enables the identifi-cation of complex structure–function relationships by analyz-ing multidimensional datasets. AI-driven approaches allow re-searchers to accelerate formulation design and guide decision-making through data-driven predictions.[21] This strategy alignswith the core principles of nanoarchitectonics, which emphasizethe deliberate, bottom-up assembly of molecular componentsinto functional nanostructures. Within this framework, AI servesas a core tool for high-throughput material screening, formula-tion optimization, and performance prediction.[22]As illustrated in Figure 3, the AI-guided design process beginsby generating diverse candidate formulations from well-curatedmaterial libraries, followed by iterative model training, predic-tion, and experimental validation.[23] These libraries encompassa broad spectrum of chemical and structural features that affectdrug loading capacity, release kinetics, targeting ability, and bio-compatibility. ML models are iteratively trained on experimen-tal datasets to uncover non-linear relationships between designvariables and delivery performance metrics. Once validated, themodels inform the design and tuning of nanocarrier structuresto achieve specific therapeutic functionalities.[24] By integratingpredictive modeling into material selection, AI facilitates a shiftfrom empirical approaches to hypothesis-driven design, promot-ing the development of more efficient and personalized drug de-livery systems.To provide a structured overview of AI applications in nanoar-chitectonics for targeted drug delivery, we categorized represen-tative approaches by their functional roles in the design workflow(Table 1). These categories include nanocarrier design, target-ing system design, and simulation or predictive modeling. EachAI approach is further classified by type, reflecting its concep-tual or operational role within the ML pipeline. Algorithms re-fer to core learning methods used for prediction, classification,or optimization (e.g., random forests (RF), support vector ma-chines (SVMs)).[25] Models represent ML architectures capableof learning complex patterns from data (e.g., deep neural net-works (DNNs), convolutional neural networks (CNNs), genera-tive models).[26] Paradigms describe broader learning strategiesthat define howmodels are trained or deployed, such as reinforce-ment learning (reward-based optimization) or federated learn-ing (distributed learning across decentralized datasets).[27] Plat-forms, such as Automated machine learning (AutoML) tools,provide integrated environments that automate model selection,Adv. Mater. 2025, 37, e10239 e10239 (3 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deFigure 2. Schematic workflow of an AI algorithm for targeted drug delivery. Created with BioRender.com.training, and tuning, allowing users to deploy robustmodels withminimal manual effort.[28]Many of the representative AI tools listed in Table 1 are avail-able as open-source platforms, which significantly enhance ac-cessibility and reproducibility in nanomedicine research. For ex-ample, SHAP enables model explainability,[29] AutoKeras sup-ports automated model development,[30] AlphaFold and RFd-iffusion are widely used for protein structure prediction anddesign,[31] and frameworks like PySyft or Flower facilitate the im-plementation of federated learning. These tools help lower tech-nical barriers and broaden participation in AI-driven nanoarchi-tectonic drug delivery design.While numerous AI approaches have been applied to nanoar-chitectonic design, their applicability, strengths, and limitationsvary significantly depending on the stage and complexity of thedrug delivery pipeline (Table 2). For instance, RF and SVMs arecommonly employed for early-stage classification tasks,[32] suchas predicting nanocarrier toxicity or ligand-binding affinity, dueto their strong performance on small to medium-sized datasetsand high interpretability. However, their ability to model non-linear, high-dimensional relationships is limited compared todeep learning methods. DNNs, including CNNs, excel at com-plex pattern recognition tasks, including structure–activity mod-eling and image-based cellular phenotyping.[33] These modelsleverage large datasets and can uncover latent features that areotherwise undetectable. Nevertheless, they are often criticizedfor their “black-box” nature, and they require extensive hyperpa-rameter tuning, high-quality data, and substantial computationalresources. Graph Neural Networks (GNNs) and transformer-based models such as AlphaFold and RoseTTAFold are highlyeffective for 3D structural predictions, particularly of proteins,peptides, and ligand–receptor interactions.[34] These models areAdv. Mater. 2025, 37, e10239 e10239 (4 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deFigure 3. AI-guided nanomaterials and nanocarrier design process for targeted drug delivery. Created with BioRender.com.increasingly applied to the design of targeting domains and bind-ing scaffolds.[30b] Despite their advantages, their complexity andreliance on high-resolution structural data can limit their acces-sibility in some research contexts. Generative models (e.g., Vari-ational Autoencoders (VAEs), Generative Adversarial Networks(GANs), RoseTTAFold diffusion (RFdiffusion) offer powerful ca-pabilities for inverse design by generating novel nanocarrier ar-chitectures with predefined features.[31b] However, these modelsare data-intensive and often require expert curation and post-processing to identify chemically viable candidates.To enhance model transparency and interpretability, explain-able artificial intelligence (XAI) tools, such as SHAP and Lo-cal Interpretable Model–Agnostic Explanations (LIME), are com-monly used to attribute importance to input features drivingmodel predictions.[29,35] Frameworks, such as knowledge graphs,support structured reasoning across complex biomedical rela-tionships and facilitate the integration of multi-scale data.[36] Fi-nally, hybrid tools that combine conventional modeling meth-ods, such as molecular dynamics (MD) simulations, with AI ac-celeration, improve both computational efficiency and predictiveperformance.[37]This classification provides a practical lens through which toevaluate and select AI tools suited to specific stages of the nanoar-chitectonic drug delivery process, supporting a more modular,scalable, and mechanistically interpretable system design, as dis-cussed in the following sections.3.1. Library-Based Nanocarrier FormulationFrom the perspective of nanoarchitectonics, nanocarrier formu-lation extends beyond material selection to the rational orga-nization of nanoscale building blocks into functional architec-tures with dynamic and integrated behavior. This approach com-bines molecular design, self-assembly, interface engineering,and external-field modulation to create adaptive drug deliverysystems.[46] To support this goal, extensive libraries of nanocar-riers have been developed, comprising a broad spectrum ofAdv. Mater. 2025, 37, e10239 e10239 (5 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deTable 1. AI approaches in nanoarchitectonics for targeted drug delivery. This table summarizes representative AI methods applied across three stagesof nanoarchitectonics for drug delivery: (1) nanocarrier design, (2) targeting system design, and (3) simulation and prediction. Each entry includes theAI approach, its functional classification (Type), a brief description, and representative applications.Stage AI Approach Type Description Example Applications Refs.1. Nanocarrier Design Deep Neural Networks (DNNs) Model Model complex relationshipsbetween formulationparameters and performancePredict nanocarrier-drugcompatibility, optimizenanoparticle synthesis[26c,d]Random Forests (RF) Algorithm Ensemble method for robustprediction and feature selectionPredict clinically relevantphysico-chemical properties ofnanoparticles[25b,c]AutoML Platforms Platform Automated ML for rapid modeldevelopmentRapid screening of materialnanotoxicity[28b]Generative Models(e.g., GANs, VAEs, etc.)Model Generate novel data or designcandidates based on learneddistributionsDesign advanced materials; Predictcrystal structures or chemicalcompositions[26g,h]Bayesian Optimization Algorithm Efficiently optimizehyperparameters orexperimental conditionsOptimize nanoparticle synthesisparameters[38]2. Targeting SystemDesignConvolutional Neural Networks(CNNs)Model Analyze structural or image-baseddataPredict ligand binding motifs,screen aptamer sequences[26e,f]Transformer Models(e.g., AlphaFold, RoseTTAFold)Model Predict 3D structures of proteinsand complexesDesign or predict the structure oftargeting peptides and antibodies[31a,39]Support Vector Machines (SVMs) Algorithm Classification and regressionmodelingPredict the binding affinity ofligands for receptors[25d–f]Reinforcement Learning (RL) paradigm Learning optimal strategiesthrough rewardsDiscover optimal ligand structuresfor specific targets[27b]3. Simulation andPredictionGaussian Process Regression(GPR)Algorithm Probabilistic modeling withuncertainty estimatesPredict binding affinity withconfidence intervals[40]Molecular Dynamics (MD)Simulation with AI AccelerationHybrid tool Simulate atomic interactions,enhanced by AI-acceleratedapproximationsPredicting efficacy of nanocarrierdesigns for cancer treatment[37]Federated Learning Paradigm Distributed learning acrossdecentralized datasetsEnable large-scale prediction ofpharmacokinetics whilepreserving data privacy[27c]Explainable Artificial Intelligence(XAI)(e.g., SHAP, LIME)Interpretability tool Interpret model decisions byfeature attributionUnderstand key factors influencingbiodistribution, ADME profiles[13c,41]Knowledge Graphs Framework Structured integration ofmulti-domain knowledgeModel relationships between drug,carrier, target, and physiologicalfactors[42]Combination ML Model Train and validate ML/DL modelswith diverse algorithms (e.g.,DNN, RF, SVM, etc.)Predict tissue distribution andtarget delivery efficiency[43]Quantitative Structure-ActivityRelationship (QSAR) ModelsAlgorithm Statistical modeling correlatingstructural descriptors withbiological activityPredict nanocarrier biodistribution,toxicity, targeting efficiency, orADME profiles[44]k-Nearest Neighbors (k-NN) Algorithm Instance-based learning methodthat classifies or predictsoutcomes based on similarity tonearest data pointsPredict a consensus RNA secondarystructure; Predict early recurrenceof disease after resection[45]building blocks including inorganic nanoparticles (e.g., goldnanoparticles, mesoporous silica, metal oxides, etc.), lipid-basedsystems (e.g., liposomes, lipid nanoparticles (LNPs)), polymer- ornucleic acid-based nanosystems, and peptide- and protein-basednanocages.[47] Each category provides distinct architectural po-tential, enabling the construction of nanocarriers with tunablesize, surface properties, and biological behavior. These materialsare engineered as integral components of orchestrated nanosys-tems, capable of exhibiting stimuli-responsive or environment-adaptive functionalities.Nanocarrier libraries are formulated with attention to bothstructural control and emergent functionality. They are builtthrough combinatorial synthesis, high-throughput screening,and systematic annotation of key physicochemical parametersAdv. Mater. 2025, 37, e10239 e10239 (6 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deTable 2.Overview of representative AImodel types commonly applied in nanoarchitectonic drug delivery design. The table summarizes the core strengths,limitations, and typical use cases of each model class, ranging from traditional machine learning (e.g., RF, SVM) to deep learning architectures (e.g.,DNN, CNN, GNN), generative frameworks, and transformer-based models. These models are selected based on their suitability for specific tasks, suchas property prediction, structural modeling, or de novo ligand design, and are referenced accordingly for further technical detail.Model Type Strengths Limitations Best Use Case Refs.Random Forests (RFs) Fast training, good for small data, andinterpretableLimited with high-dimensional orsequential dataProperty prediction, feature selection [32a]Support Vector Machines (SVMs) Accurate classification, effective forsmall dataHard to scale to large datasets,kernel selectionBinding affinity classification [32b,c]Deep Neural Networks (DNNs) Powerful nonlinear modeling Requires large data, lessinterpretableStructure–activity prediction [33b,d]Convolutional Neural Networks(CNNs)Excellent for image-based analysis Less effective for tabular data Phenotypic screening, imageclassification[33a,c]Graph Neural Networks (GNNs) Captures complex molecular graphs Complex to train, needsstructured inputProtein-ligand modeling, bindingprediction[34a,d]Generative Models (GAN, VAE,DDPM)Enables inverse design, creativeexplorationRisk of invalid outputs, needscurationNovel ligand or carrier design [31b,34b]Transformers State-of-the-art for structuralpredictionHigh resource demands, limitedgeneralizabilityProtein structure or interactionprediction and design[31a,34c]such as size, zeta potential, drug loading efficiency, release pro-files, targeting capability, and biocompatibility.[48] The resultingdatasets provide a critical foundation for AI-driven selection andoptimization. The effectiveness of AI-assisted design dependssignificantly on the intrinsic characteristics of each nanocarrierplatform. For example, inorganic systems provide high struc-tural stability and well-defined surface properties, which enableprecise ligand functionalization and compatibility with imagingapplications.[49] Lipid-based carriers are clinically validated, bio-compatible, and highly efficient for drug encapsulation, makingthem well-suited for AI-based modeling of membrane dynamicsand release kinetics. Polymeric nanocarriers offer tunable degra-dation rates, mechanical flexibility, and diverse drug–polymer in-teractions, which can be leveraged by ML to optimize spatiotem-poral release profiles.[50] Meanwhile, nucleic acid-based architec-tures and protein nanocages offer genetically programmable fea-tures that allow AI to assist in folding prediction, sequence opti-mization, and immunogenicity evaluation.[51]In addition to conventional nanocarrier types, AI-assisted de-sign is increasingly being applied to more complex delivery plat-forms such as multi-stage responsive systems and multifunc-tional synergistic carriers.[52] These advanced architectures areengineered to perform sequential or simultaneous functions,including targeted accumulation, controlled drug release in re-sponse to specific stimuli, and integrationwith diagnostic or ther-apeutic modalities.[5] The design space for such systems is inher-ently high-dimensional, making them well-suited for data-drivenmodeling and optimization. Machine learning algorithms, in-cluding deep learning and reinforcement learning, can be usedto predict interdependent behaviors among components, opti-mize spatiotemporal response profiles, and support the inversedesign of multifunctional carriers tailored for complex diseasecontexts.[53]Recognizing the distinct properties of each nanocarrier classis essential for developing AI models that are appropriatelymatched to the design complexity and intended function. As aresult, nanocarrier formulation libraries serve not only as a foun-dation for material discovery but also as context-specific envi-ronments for AI-based inverse design, optimization, and perfor-mance prediction. Among AI techniques, VAEs and GANs haveemerged as powerful generative models for the inverse design ofinorganic solid materials.[54] ML models trained on nanocarrierlibraries can uncover hidden structure–property–function rela-tionships that inform the rational design of next-generation deliv-ery systems. Rather than serving as static material catalogs, for-mulation libraries function as dynamic design spaces. In thesespaces, each nanocarrier is treated as an architectural unit whoseform and function can be computationally predicted, experimen-tally tuned, and systematically improved.3.2. Iterative ML Model DevelopmentWhile earlier sections have emphasized predictive modeling forformulation optimization, recent advances have expanded therole of AI in nanoarchitectonics from performance estimation toautonomous design generation. Building on data from nanocar-rier libraries, machine learning models are developed throughiterative training cycles that integrate experimental feedback.These models predict key delivery metrics such as encapsula-tion efficiency, circulation half-life, tissue-specific accumulation,and therapeutic efficacy based on design parameters.[12a,55] Algo-rithms such as random forests, SVMs, gradient boosting, andDNNs are commonly used to capture nonlinear relationships be-tween formulation features and biological outcomes (Table 1).In nanoarchitectonics, model construction extends beyondconventional regression or classification tasks. AI functions as acomputational orchestrator, harmonizing design elements frommolecular building blocks to complete nanosystems by learninghow small-scale attributes, such as surface ligands, charge distri-bution, or core structure, translate into emergent behaviors liketumor-specific accumulation or endosomal escape.Recent breakthroughs in generative and structure-predictive AI have enabled the bottom-up design of molecularAdv. Mater. 2025, 37, e10239 e10239 (7 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deFigure 4. Representative examples of model construction for AI-driven nanomedicine design. A) Generative design of cyclic symmetric protein nanos-tructures using a Monte Carlo sampling approach guided by AlphaFold2 (AF2) structure predictions and ProteinMPNN-based sequence refinement.A random sequence is iteratively optimized to fit target symmetry and structural confidence. Reproduced with permission.[56] Copyright 2022, AAAS.B) Fine-tuning of bifaceted protein nanoparticle subunits using RFdiffusion. Backbone structures connecting functional motifs are generated de novo,followed by sequence design with ProteinMPNN and structural validation via AF2. Reproduced under the terms of the CC-BY Creative Commons Attri-bution 4.0 International License.[57] Copyright 2024, bioRxiv. C) Iterative optimization of polymeric nanoparticle formulations via a closed-loop system.Microfluidic-assisted nanoprecipitation produces formulation variants, which are screened using high-content imaging (HCI). AnMLmodel is trained torelate formulation variables to biological responses, enabling active learning and rapid design cycles. Reproduced under the terms of the CC-BY CreativeCommons Attribution 3.0 Unported License[25c] Copyright 2024, RSC. D) A three-stage nanoparticle discovery pipeline leveraging self-supervised andsupervised learning for ionizable LNP design. The AGILE framework pre-trains on virtual libraries, fine-tunes with high-throughput data, and deployspredictive models for candidate ranking in mRNA delivery applications. Reproduced under the terms of the CC-BY Creative Commons Attribution 4.0International License.[58] Copyright 2024, Springer Nature. Created with BioRender.com.architectures with predefined structural and functional proper-ties. For instance, Markov chain Monte Carlo (MCMC)-drivenAlphaFold2 pipelines can generate novel protein nanostructuresstarting from minimal constraints such as cyclic symmetry andchain length, with sequence optimization by Protein Message-Passing Neural Network (ProteinMPNN) ensuring stable foldingand functional interfaces (Figure 4A).[56] These models not onlyclassify or regress known formulations but also create entirelynew molecular architectures with defined structural motifsand self-assembly capabilities, embodying the principles ofbottom-up functional design. Similarly, RFdiffusion, a denois-ing diffusion probabilistic model, has been applied to designbifunctional protein nanoparticles by generating backbonestructures that bridge functional domains, followed by closed-loop sequence refinement and verification using AlphaFold2(Figure 4B).[57]Adv. Mater. 2025, 37, e10239 e10239 (8 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deThese approaches demonstrate how AI can act as a digital ar-chitect, combining structural biology, computational design, andmaterials science to accelerate nanoarchitectonic innovation. Fur-ther integration with active learning strategies enables ML mod-els to prioritize high-value data points for experimental testing,closing the loop between computational prediction, nanoscalesynthesis, and biological validation. This approach not only en-hances efficiency but also advances the vision of self-refining,adaptive design systems in AI-driven nanomedicine.3.3. Optimization of Nanocarrier Delivery PerformanceIntegrated experimental-AI pipelines have emerged that com-bine formulation, high-throughput screening, and ML-guidediteration. These systems reflect the principles of nanoarchitec-tonics by establishing tightly coupled loops of design, construc-tion, and evaluation that drive emergent, system-level behavior.This approach enables data-driven refinement of nanocarrier sys-tems, leading to improved targeting, bioavailability, and thera-peutic performance.One representative approach utilizes microfluidics-assistednanoprecipitation to generate nanoparticles with controlled vari-ations in composition and structure, forming a dynamic library ofcandidate formulations (Figure 4C).[25c] These particles are thenevaluated using high-content imaging (HCI) to extract pheno-typic data such as cellular uptake, intracellular trafficking, andcytotoxicity. The resulting data feeds into an ML model, whichlearns the relationship between formulation variables and bio-logical outcomes. The model then suggests new formulationspredicted to improve desired properties, thereby guiding thenext experimental cycle in an active learning loop. This strategyhas been successfully applied to discover poly(lactide-co-glycolicacid)(PLGA)- polyethylene glycol (PEG) nanoparticles with en-hanced uptake in breast cancer cells and holds promise for di-verse nanocarrier platforms.Advanced AI frameworks such as the AI-guided ionizable lipidengineering (AGILE) platform, based on a GNN architecture,implement a modular, staged approach that mirrors nanoarchi-tectonic construction (Figure 4D).[58] In this system, a virtual li-brary of ionizable lipids is first used to pretrain a DL model ina self-supervised manner. This is followed by supervised fine-tuning using high-throughput experimental data to capture real-world performance trends. The trained model is then deployedto predict optimal lipid compositions for mRNA delivery viaLNPs. By learning the complex structure–function relationshipsthat govern transfection efficiency, membrane interaction, andimmunogenicity, AGILE accelerates the identification of high-performance nanocarriers across vast chemical design spaces.Through such frameworks, AI-driven optimization transcendsconventional screening by enabling a convergence of computa-tion and experimentation. Complex design objectives, such astarget selectivity, stimuli-responsiveness, ormulti-drug coordina-tion, can be addressed through intelligent iteration within a mul-tidimensional design space. A real-world demonstration of thisapproach is the development of LNP-based mRNA COVID-19vaccines by Pfizer-BioNTech andModerna.[59] In these platforms,AI and ML techniques were employed to optimize the composi-tion of ionizable lipids, PEG-lipids, and cholesterol derivatives,which are crucial for ensuring efficient mRNA encapsulation,endosomal escape, and reduced immunogenicity. Computationalscreening methods, including Bayesian optimization and GNNs,facilitated the rapid evaluation of vast lipid libraries to predictcandidate formulations with favorable biophysical properties andhigh transfection efficiency.[60] These AI-accelerated pipelinessignificantly shortened the formulation timeline and improvedin vivo performance, enabling rapid clinical translation. How-ever, this process still required extensive experimental validation,adaptation toGoodManufacturing Practice (GMP) protocols, andregulatory engagement to address safety, scalability, and batchconsistency, which underscores the real-world hurdles that existeven in successful translation cases.[61] This paradigm has alsobeen extended to siRNA-based nanocarrier systems, with sev-eral candidates designed using predictive modeling tools advanc-ing into early-phase clinical trials, highlighting their translationalpotential. Nevertheless, translation of AI-designed siRNA carri-ers faces challenges such as interpatient variability in immuneresponses, difficulty in predicting long-term pharmacokinetics,and regulatory uncertainty regarding model interpretability andreproducibility.In a typical implementation, DNNs are trained on datasetscomprising nanoparticle descriptors such as size, surface charge(zeta potential), PEG density, hydrophilicity, ligand valency, andencapsulation efficiency. These features are paired with experi-mental outcomes, including cellular uptake rates, biodistributionprofiles, or in vivo therapeutic response. GNNs are particularlysuited for modeling polymeric or lipid structural relationshipsand can predict self-assembly behavior or membrane interactionpotential. Trained models are used to rank or generate candi-date formulations with optimized delivery profiles, which are ex-perimentally validated and used to refine future predictions in aclosed-loop system.Collectively, thesemodel-based strategies illustrate how AI cannot only interpret but also generate and optimize nanocarrier sys-tems with minimal human intervention. The synergy betweendesign algorithms, predictive analytics, and experimental feed-back loops is reshaping the landscape of nanomedicine devel-opment, enabling faster, smarter, and more personalized drugdelivery platforms. These AI-guided optimization strategies setthe stage for a more structured approach to smart nanocarrierdesign. In the following section, we present a three-phase frame-work that integrates thesemethodologies into a coherent strategyfor precision targeting.4. A Three-Step Strategy for AI-Driven SmartNanoarchitectonicsThis section introduces a structured, AI-enabled frameworkfor the rational design of advanced targeted drug deliverysystems. The proposed strategy consists of three interlinkedphases that incorporate AI at every stage of the design pipelineto enhance specificity, adaptability, and therapeutic efficacy(Figure 5). The first phase focuses on molecular target identifi-cation through bioinformatic profiling and ML-based analysis ofhigh-dimensional omics data. The second phase involves predic-tive surface engineering, where ML models guide the structuraltuning and functionalization of nanocarriers to achieve selectivetargeting and immune compatibility. The third phase centers onAdv. Mater. 2025, 37, e10239 e10239 (9 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deFigure 5. AI-driven prediction and optimization of smart targeted drug delivery. Created with BioRender.com.delivery optimization, employing in silico simulations and data-informed pharmacokinetic modeling to refine biodistribution,binding kinetics, and systemic performance. Collectively, thesesteps provide a mechanistically grounded and data-driven frame-work that aligns with the principles of nanoarchitectonics andsupports adaptive design across molecular, structural, and phys-iological scales.Each phase comprises distinct but interconnected computa-tional and experimental tasks. Rather than following a strictlylinear sequence, the workflow operates iteratively: findingsfrom surface functionalization or delivery modeling can informearlier-stage decisions, including target prioritization and ligandselection. Such feedback-driven refinement increases respon-siveness to biological complexity and supports context-sensitivedesign decisions. By reflecting the self-regulating nature of liv-ing systems, the recursive architecture exemplifies nanoarchitec-tonic principles that emphasize functional integration, scalabil-ity, and structural adaptability.AI integration across the design pipeline not only enhancesprediction accuracy but also accelerates translation from in sil-ico models to in vivo performance. The sections that follow ex-amine each phase in detail, including (Section 4.1) target iden-tification, (Section 4.2) precision surface engineering, and (Sec-tion 4.3) dynamic delivery optimization, highlighting howAI andML methodologies enable hypothesis generation, feature extrac-tion, system-level learning, and iterative model refinement tosupport the creation of next-generation, intelligent drug deliverysystems.4.1. Rational Target Identification through Bioinformatic ProfilingThe initial phase of AI-driven targeted drug delivery centerson selecting biologically meaningful targets, which is a deci-sion that defines the entire trajectory of nanocarrier design.Traditional target identification relies on experimentally vali-dated markers or phenotypic screening, which are often labor-intensive and limited in scope. However, recent advances indata science have enabled the analysis of high-dimensional,multi-omics datasets, such as genomic, transcriptomic, pro-teomic, and metabolomic profiles, to uncover previously un-recognized biomarkers and disease-specific pathways.[62] Tech-niques such as unsupervised clustering, principal componentanalysis, and feature selection algorithms allow for patientstratification and identification of conserved molecular pat-terns suitable for targeting, thereby enhancing therapeuticprecision.Single-cell RNA sequencing (scRNA-seq) has emerged as apowerful tool to resolve human disease heterogeneity at cel-lular resolution, particularly when conventional models areinsufficient.[63] By applying scRNA-seq to patient-derived sam-ples, researchers can identify disease-associated gene expres-sion signatures and determine which genes or proteins mightserve as optimal therapeutic targets. For example, analysisof patient-derived samples in drug reaction with eosinophiliaand systemic symptoms/drug-induced hypersensitivity syn-drome (DIHS/DRESS) revealed distinct immune cell signa-tures and upregulation of the JAK-STAT signaling pathway,pinpointing it as a potential therapeutic target (Figure 6A).[64]This case exemplifies how bioinformatic profiling via scRNA-seq can guide rational target selection for immune-mediateddiseases.Natural language processing (NLP) tools further enhancetarget discovery by mining scientific literature and clinicaltrial databases to contextualize candidate genes within diseaseontologies.[65] When combined with biological network analy-ses, such as protein–protein interaction mapping or pathway en-richment models, these tools strengthen target prioritization byAdv. Mater. 2025, 37, e10239 e10239 (10 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deFigure 6. Representative examples for rational target identification through bioinformatic profiling. A) Single-cell RNA sequencing (scRNA-seq) analysisenables high-resolution profiling of patient-derived samples, revealing disease-specific cellular states and transcriptomic signatures. In the exampleshown, scRNA-seq of skin and blood samples from patients with refractory DiHS/DRESS identified an enrichment of keratinocytes and immune cellsubpopulations, highlighting potential cellular targets. Reproduced with permission.[64] Copyright 2020, Springer Nature. B) PNAbind, a GNN model,predicts protein binding sites for peptide nucleic acids (PNAs) based on structural and physicochemical properties, allowing identification of therapeuti-cally relevant interaction residues. Reproduced under the terms of the CC-BY Creative Commons Attribution 4.0 International License. [67] Copyright 2024,Springer Nature. C) MaSIF (Molecular Surface Interaction Fingerprinting) detects targetable surface patches and evaluates interaction complementaritythrough DL. MaSIF-site identifies interface-prone regions, while MaSIF-seed screens a database of millions of patches to rank potential binding partners,accelerating rational binder design. Reproduced under the terms of the CC-BY Creative Commons Attribution 4.0 International License.[68] Copyright2023, Springer Nature. Created with BioRender.com.identifying high centrality nodes with therapeutic relevance.MarkerGeneBERT, an NLP-based model trained on curatedbiomedical corpora, automatically extracts cell marker genesfrom single-cell sequencing studies, improving annotation accu-racy and accelerating scRNA-seq interpretation.[66] This exempli-fies how NLP methods can systematically augment early-stagetarget validation, forming part of a broader informatics pipelinethat enhances the specificity and robustness of nanocarrier tar-geting strategies.Once a target protein has been identified for therapeutic or di-agnostic purposes, it is essential to assess its druggability andidentify optimal binding interfaces. DL tools such as PNAbindemploy GNNs to predict protein binding sites for peptide nucleicacids (PNAs) by analyzing structural and physicochemical fea-tures (Figure 6B).[67] Similarly,Molecular Surface Interaction Fin-gerprinting (MaSIF) uses DL to identify interface-prone surfacepatches and evaluate their compatibility with potential binders(Figure 6C).[68] The MaSIF-site module predicts regions likely toform buried protein–protein interfaces, while MaSIF-seed com-pares these patches to a database of over 400 million interactionmotifs, aligning and rescoring the top candidates to identify vi-able seed structures. Together, these tools provide a robust andscalable platform for structure-based ligand design using only thestructural information of the target protein.In parallel, bioinformatic profiling should also account fordelivery-related biological constraints. These include hetero-geneous receptor expression across patient populations, off-target cell interactions, stromal barriers, elevated interstitialpressure, and the presence of immune cells that can rapidlyclear nanoparticles.[2b] Identifying these obstacles early helpsinform downstream design decisions such as surface modi-fication or choice of delivery route. For example, physiologi-cal barriers, including plasma opsonization, tumor stroma den-sity, and interstitial fluid pressure, can influence nanoparti-cle accumulation and distribution.[69] Furthermore, transcrip-tomic and proteomic analyses can be leveraged not only topinpoint overexpressed surface receptors (e.g., HER2, EGFR,CD44), but also to uncover local microenvironmental stim-uli such as hypoxia, acidic pH, or redox imbalance. Thesefeatures can serve as activation triggers for smart, stimuli-responsive nanocarriers.[70] Incorporating both target recogni-tion and biological barriers into the profiling stage enablesmore accurate and context-specific design of AI-guided deliverysystems.Adv. Mater. 2025, 37, e10239 e10239 (11 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deFigure 7. Representative examples of ML-guided surface engineering for nanocarrier targeting precision. A) RFdiffusion, a generative protein designmodel built upon RoseTTAFold, enables the de novo creation of receptor-binding proteins through iterative denoising of random structural inputs. Themodel generates folded backbones with user-defined topologies, facilitating high-specificity ligand design. Reproduced under the terms of the CC-BYCreative Commons Attribution 4.0 International License. [31b] Copyright 2023, Springer Nature. B) Application of RFdiffusion to design a binder againstthe neurotoxin 𝛼-cobratoxin. Starting from randomized residue placement near target 𝛽-strands, themodel iteratively refines structures to produce a high-affinity binding protein. In vitro validation confirms strong neutralization comparable to monoclonal antibodies. Reproduced under the terms of the CCBY-NC-ND Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.[72] Copyright 2025, Springer Nature. C) RaptGen,a generative model trained on SELEX datasets, designs aptamers by encoding motif features into a latent space using a profile HMM decoder. Themodel predicts activity of novel sequences and uses Bayesian optimization to guide aptamer refinement across varying nucleotide lengths. Reproducedunder the terms of the CC-BY Creative Commons Attribution 4.0 International License.[38] Copyright 2022, Springer Nature. D) Surface ligand densityoptimization on nanocarriers. AI-guided modeling highlights the importance of matching ligand presentation to receptor density for maximizing cellularinternalization, as demonstrated in ErbB2-targeting systems. Design variables include ligand type, spatial distribution, linker length, and surface charge.Reprinted with permission.[74] Copyright 2020, American Chemical Society. Created with BioRender.com.4.2. ML-Guided Surface Engineering for Precision TargetingFollowing target identification, selective ligand design is essentialto ensure precise and efficient drug delivery. These strategies aregenerally categorized into passive and active targeting. Amongthem, active targeting, which is based on ligand–receptor inter-actions, plays a key role in precision medicine, offering advan-tages over passive strategies like the enhanced permeability andretention (EPR) effect or liver-specific uptake by LNPs, which arelimited to certain physiological contexts.[71]Recent advances in DL have enabled the de novo design ofhigh-affinity targeting moieties, including small molecules, pep-tides, and antibodies. One example is RFdiffusion, a genera-tive protein design tool built on the RoseTTAFold architecture.Fine-tuned for structural denoising tasks, it generates foldedprotein backbones with defined topologies by iteratively refin-ing random sequences. RFdiffusion has been used to designreceptor-binding proteins, enzyme scaffolds, and therapeuticmotifs (Figure 7A).[31b] Notably, it produced a high-affinity binderagainst the neurotoxin 𝛼-cobratoxin, with in vitro validation con-firming strong neutralization comparable to that of monoclonalantibodies (Figure 7B).[72] Simultaneously, RaptGen, a generativemodel trained on SELEX (Systematic Evolution of Ligands byEXponential enrichment) datasets, designs aptamer sequencesby embedding motif features into a latent space and optimizingcandidates through Bayesian approaches, producing short, func-tional sequences with high binding activity (Figure 7C).[38]Once designed, targeting ligands are conjugated to nanocar-rier surfaces, where spatial configuration, ligand density, andsurface chemistry critically influence cellular targeting anduptake.[73] For instance, Figure 7D illustrates how optimizedligand-to-receptor ratios enhance internalization in ErbB2-targeting systems, highlighting the importance of nanoscaleprecision.[74] The design space for nanoparticle surfaces is com-plex, involving variables such as ligand type, linker length, spa-tial density, hydrophilic–hydrophobic balance, and overall sur-face charge.[75]Stealth functionality is another critical aspect of surface de-sign, as it determines nanoparticle circulation time and immuneevasion. Without effective surface shielding, nanocarriers areAdv. Mater. 2025, 37, e10239 e10239 (12 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.derapidly cleared by the mononuclear phagocyte system. Strategiessuch as PEGylation and zwitterionic coatings are commonly em-ployed to reduce opsonization and enhance systemic stability.[76]Designing such surfaces involves tuning parameters like charge,hydrophilicity, steric effects, and ligand presentation. This task ischallenging due to the vast design space and the dynamic na-ture of biological environments. ML provides a powerful toolto streamline this process. ML models trained on in vivo datacan predict how variations in PEG density, chain length, or spa-tial arrangement influence stealth performance and targetingefficiency.[4a,77]In addition to stealth optimization, AI is increasingly appliedto fine-tune other biophysical surface properties. ML algorithmscan predict ligand–receptor binding affinity and simulate cel-lular uptake across various surface configurations. Techniquessuch as genetic algorithms and Bayesian optimization enablemulti-parameter refinement under physiological constraints, al-lowing for precise control over ligand orientation, multivalency,and surface charge to enhance specificity and reduce off-targeteffects.[26f,44c,78]Recent studies have also applied ML methods like randomforests, k-nearest neighbors (k-NN), and SVM to predict nanopar-ticle dispersibility in solvents, using molecular descriptors suchasHansen solubility parameters, COSMO-based quantum chem-ical properties, and MACCSKeys fingerprints.[79] These tools areincreasingly important in streamlining formulation and scalingin industrial contexts. Together, AI-enabled surface engineeringstreamlines the development of dynamic and adaptive nanocar-rier systems, advancing the field toward more precise and cus-tomizable drug delivery platforms.4.3. Predictive Modeling of Binding and Delivery KineticsWhile surface design determines the physical interface for cellu-lar recognition, the dynamic interactions involved in drug deliv-ery must be evaluated computationally. This section focuses onpredictivemodeling of nanocarrier behavior across biological sys-tems, including molecular recognition such as ligand–receptorbinding, as well as broader processes like membrane transloca-tion, intracellular trafficking, and endosomal escape.[80] In silicoapproaches support early-stage screening by identifying promis-ing candidates before laboratory validation, which reduces devel-opment time and resource requirements.At the molecular scale, recent progress in DL has significantlyimproved the prediction of binding interactions. For instance,SurfDock combines protein sequences, 3D structural graphs,and surface features within a generative diffusion framework topredict ligand–receptor binding poses with high accuracy, out-performing traditional docking algorithms (Figure 8A).[81] Si-multaneously, MD simulations remain essential for understand-ing the binding behavior of ligands. They provide atomistic in-sights into stability, energy landscapes, and the effects of sol-vent environments or external stimuli. These simulations areparticularly helpful in refining ligand orientation and assess-ing how structural flexibility or multivalency influences bindingefficiency.[37]To model larger-scale interactions, coarse-grained methodssuch as the Martini force field are used to simulate nanoparti-cle interactions with cell membranes, including adhesion, en-docytosis, and release into the cytoplasm. Potential of meanforce (PMF) simulations further quantify energy barriers asso-ciated with membrane penetration and molecular binding, in-forming the thermodynamic feasibility of specific delivery routes(Figure 8B).[82]AI-guided simulations are increasingly integrated with tra-ditional modeling techniques to improve efficiency and accu-racy. These hybrid methods can accelerate sampling, learn po-tential energy surfaces, and identify stable configurations withgreater precision. For example, AlphaFold3 exhibits remark-ably high accuracy in predicting complex biomolecular struc-tures, including proteins, nucleic acids, and ligand assemblies(Figure 8C).[31a] These AI-driven structural prediction tools sup-port structure-based nanocarrier design and enable more refinedmodeling of interactions between ligands and dynamic biolog-ical interfaces. AI-guided force field tuning also helps capturesubtle interactions that are difficult to model with conventionalparameters.In the case of LNP-based systems, in silico models are beingused to predict transfection efficiency by correlating structuralproperties with functional outcomes (Figure 8D).[58] These pre-dictions are often supported by experimental studies measuringmRNA expression, cellular uptake, or protein production. Thiscontinuous exchange between in silico predictions and labora-tory validation enables rapid refinement of LNP formulations forgene delivery applications.[83]Importantly, accurate prediction of binding and delivery ki-netics also requires consideration of complex biological barri-ers that influence nanocarrier transport and function.[84] Interac-tions with plasma proteins, such as albumin and immunoglobu-lins, can lead to opsonization and formation of a protein corona,which significantly alters the pharmacokinetic profile and tar-geting efficiency of nanoarchitectures. Likewise, the accessibil-ity and expression pattern of cellular surface receptors at thedisease site can impact ligand–receptor recognition, internaliza-tion, and tissue specificity. Other microenvironmental compo-nents, including the extracellular matrix, glycocalyx, and intersti-tial fluid pressure,may impede diffusion or promote non-specificuptake.[85] Incorporating these parameters into AI and simula-tion frameworks is crucial to achieving physiologically relevantpredictions and to supporting the rational design of robust andtranslatable delivery systems.[86]Beyond physics-based modeling, data-driven approaches suchas quantitative structure–activity relationship (QSAR) modelingand ML algorithms constitute a critical factor. While QSAR isnot inherently AI-based, it is increasingly combined with ML tocorrelate physicochemical features, such as size, surface charge,and hydrophobicity, to biological responses, including cellularuptake, toxicity, and biodistribution. By learning from these rela-tionships, QSAR and AI-based models can prioritize virtual can-didates that are likely to exhibit favorable performance in vitroand/or in vivo.[87] For instance, QSAR models have been appliedto predict tumor accumulation of polymeric nanoparticles basedon surface chemistry and mechanical properties.[88] Classifica-tion algorithms, such as SVMs and random forests, have distin-guished between high- and low-performing formulations, whileDL models have predicted cytotoxicity and trafficking behaviorsin lipid-based carriers with notable accuracy.[89]Adv. Mater. 2025, 37, e10239 e10239 (13 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deFigure 8. Computational strategies for in silico optimization of nanocarrier binding and delivery dynamics. A) SurfDock, a deep-learning-based methodthat integrates protein sequences, 3D structural graphs, and surface-level features within an equivariant architecture to predict ligand–receptor bindingposes. Reproduced with permission.[81] Copyright 2025, Springer Nature. B) Potential of mean force (PMF) profiles derived from umbrella samplingsimulations to quantify energy barriers associated with membrane translocation and molecular binding. Reproduced under the terms of the CC BY-NC-ND Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.[82] Copyright 2025, Springer Nature. C) AlphaFold3model capable of high-accuracy prediction of protein–ligand, protein–nucleic acid, and antibody–antigen interactions within a unified DL architecture,surpassing previous specialized tools in structural accuracy. Reproduced under the terms of the CC-BY Creative Commons Attribution 4.0 InternationalLicense.[31a] Copyright 2024, Springer Nature. D) In silico prediction of lipid nanoparticle (LNP) transfection efficiency based on structure–functionrelationships, forming an iterative design–validation loop with experimental readouts such as mRNA expression and protein production. Reproducedunder the terms of the CC-BY Creative Commons Attribution 4.0 International License.[58] Copyright 2024, Springer Nature. Created with BioRender.com.Model interpretability is an important consideration. Meth-ods such as SHAP provide insights into which molecular orstructural features most strongly influencemodel predictions.[90]These tools not only enhance transparency but also help guidemechanistic understanding and experimental design. Addition-ally, pharmacokinetic and pharmacodynamic (PK/PD) model-ing can be integrated with AI to optimize dose schedules, im-prove delivery efficiency, and tailor therapies to specific dis-ease contexts.[91] AI-powered ADME (Absorption, Distribution,Metabolism, and Excretion) simulations further provide mech-anistic insights into in vivo drug behavior, informing the ra-tional design of delivery systems and reducing systemic toxi-city through improved targeting.[13c,41,44b] These computationalstrategies support precise modeling of nanocarrier interactionsand facilitate the rational design of adaptive delivery systemswithin the broader framework of nanoarchitectonics.Furthermore, targeted delivery can be conceptualized as amultiscale cascade encompassing organ-level biodistribution,cell-specific targeting, and intracellular trafficking to functionalcompartments such as the cytosol or nucleus.[4a] AI modelscontribute at each of these levels by predicting systemic dis-tribution from formulation attributes, guiding ligand selec-tion for receptor-mediated endocytosis, and simulating mem-brane translocation or endosomal escape through in silicoapproaches.[44b,92] Key performance metrics, including targetingefficiency, cellular uptake rate, binding affinity, and endosomalescape probability, serve both as design objectives and as feed-back variables for iterative refinement of AI models.Adv. Mater. 2025, 37, e10239 e10239 (14 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.de5. Challenges and Future OutlookThe integration of AI with nanoarchitectonics offers unprece-dented opportunities for precision drug delivery but also presentsseveral foundational challenges. Among these, data quality,model interpretability, regulatory readiness, and ethical de-ployment stand out as critical areas that require coordinatedsolutions.[93]Data Scarcity: The development of accurate and generalizableAI models in nanomedicine is hindered by a lack of high-qualityand standardized datasets. Current data are often fragmented,inconsistently reported, and lack harmonization across key pa-rameters such as surface chemistry, pharmacokinetics, and ther-apeutic outcomes. There is also a pronounced gap between pre-clinical and clinical datasets, which complicates model transfer-ability and real-world validation. To address these issues, FAIR(Findable, Accessible, Interoperable, and Reusable) data princi-plesmust be considered through community-driven curation ini-tiatives. Federated learning offers a compelling solution by al-lowing decentralized model training across multiple institutionswithout compromising data privacy or ownership, thereby en-abling secure and large-scale collaboration.Model Interpretability: Although DL models demonstratestrong predictive capabilities in nanomedicine, their internaldecision-making processes are often difficult to interpret, espe-cially in complex biomedical systems. This limited transparencypresents challenges for clinical translation and regulatory ap-proval, where understanding how input features influencemodeloutputs is essential for error detection, hypothesis generation,and risk assessment. To address this issue, XAI frameworks andinterpretable model architectures are increasingly used to clarifyhow specific nanocarrier attributes, such as particle size, charge,or morphology, contribute to biological responses like biodistri-bution and immune evasion. Techniques such as saliency map-ping, feature attribution (e.g., SHAP values), and counterfactualanalysis enable researchers to visualize feature relevance andextract mechanistic insight from trained models. For example,SHAP can quantify the relative influence of design parameterssuch as zeta potential, PEG density, and particle diameter on pre-dicted biodistribution profiles. These insights not only supportrational nanocarrier optimization but also improve model trans-parency in the context of regulatory review.In parallel, federated learning architectures are gaining atten-tion as a means to enhance model reproducibility and gener-alizability. These frameworks allow multiple institutions to col-laboratively train AI models on decentralized yet harmonizeddatasets, supporting robust validation across diverse clinical en-vironments while maintaining data privacy and compliance withregulatory standards.[94]Regulatory Hurdles: Current regulatory frameworks were notdesigned to accommodate adaptive, continuously learning ther-apeutic platforms. Unlike conventional drug formulations, AI-driven systems evolve based on newly acquired data, whichchallenges existing standards for validation, reproducibility, andsafety assessment.Recognizing these complexities, regulatory bodies are begin-ning to issue AI-specific guidance. For example, the U.S. Foodand Drug Administration (FDA) released a draft guidance in2024 outlining key considerations for the use of AI in drugand biological product development. This includes expectationsfor data quality, model transparency, explainability, generaliz-ability, and lifecycle management.[95] Similarly, the EuropeanMedicines Agency (EMA) published its Network Data SteeringGroup (NDSG) Workplan 2025–2028, emphasizing the integra-tion of AI governance into the medicinal product lifecycle andaligning with broader regulations such as the EU AI Act andGDPR. These efforts reflect a global shift toward regulatory in-novation in response to AI-driven technologies.[96]Continued progress will require adaptive guidelines, regu-latory sandboxes, and standardized validation pipelines. Inter-disciplinary working groups that include experts in regulatoryscience, computational modeling, clinical pharmacology, andethics can help establish risk-informed pathways that ensureboth safety and innovation. The growing use of AI platforms innanomedicine also presents new risks in intellectual propertyprotection. Premature disclosure of AI-generated designs, am-biguous inventorship, or third-party platform data rights maycompromise novelty or ownership. These risks highlight theneed for legal and institutional safeguards to protect innovationwhile ensuring responsible development.[97]Ethical Concerns: Algorithmic bias, data security, andinformed consent are especially relevant in AI-enablednanomedicine. Models trained on non-representative or histor-ically biased datasets may reinforce existing health disparities,resulting in inaccurate risk stratification, suboptimal drugrecommendations, or even harmful clinical decisions for un-derrepresented populations. Furthermore, sensor-integratedand feedback-driven systems raise concerns around continuoussurveillance, loss of autonomy, and the adequacy of one-timeconsent in the context of dynamic, adaptive technologies. Theserisks call for ethical design principles to be embedded early indevelopment through frameworks such as ethics-by-design andalgorithmic auditing. Transparent data governance, inclusivedesign practices, and stakeholder engagement are required formaintaining equity and accountability as technology evolves.[98]Strategic Directions: To address these multidimensional chal-lenges, innovationmust occur at both technical and systemic lev-els. The development of interpretable and context-aware AI mod-els is critical. These models should not only deliver predictiveaccuracy but also contribute to hypothesis generation, formula-tion optimization, and therapeutic decision-making. Advances infederated learning, synthetic data generation using models suchas GANs or diffusion algorithms, and multimodal data integra-tion that combines omics, imaging, and biophysical inputs willstrengthen model robustness and applicability across diverse bi-ological contexts. AI-driven nanocarrier design platforms, whichintegrate autonomous laboratories, robotic synthesis, and closed-loop experimentation, offer opportunities to accelerate the itera-tive design process. These systems can efficiently couple simula-tion and real-time experimentation, reducing development timeand improving candidate prioritization (Figure 9)The emergence of real-time optimized delivery systems rep-resents a major leap forward. These platforms use biosensorfeedback to dynamically adjust dosing, targeting, and release inresponse to patient-specific signals. Such adaptability enablesunprecedented therapeutic precision. Additionally, the conver-gence of AI withmultimodal nanomedicine, which integrates ge-nomics, transcriptomics, imaging, and biomechanical data, willAdv. Mater. 2025, 37, e10239 e10239 (15 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.deFigure 9. Conceptual framework summarizing the key challenges and future directions in AI-integrated nanoarchitectonics for targeted drug delivery.Created with BioRender.com.support the creation of multifunctional nanocarriers capable ofsimultaneous diagnosis and treatment. Achieving this vision re-quires regulatory and ethical frameworks that are as adaptive andintegrated as the technologies they aim to govern. By viewing cur-rent challenges not as barriers but as design constraints, the fieldcan foster a more resilient and innovative ecosystem. In this con-text, AI-integrated nanoarchitectonics has the potential to shapea new paradigm in precisionmedicine that is both intelligent andethically grounded.AcknowledgementsThis research was supported by a grant (RS-2024-00331900) from theMin-istry of Food and Drug Safety in 2024. This work was supported by theNational Research Foundation of Korea (NRF) grant funded by the Koreagovernment (MSIT) (No. 2023R1A2C3005731). [Correction added on Au-gust 14, 2025, after first online publication: Acknowledgement Section hasbeen updated + Figures 1–9 has been replaced.]Open access publishing facilitated by University of Wollongong, as partof the Wiley - University of Wollongong agreement via the Council ofAustralian University Librarians.Conflict of InterestThe authors declare no conflict of interest.Keywordsartificial intelligence, in silico optimization, machine learning, nanocarrierdesign, nanomedicine, targeted drug deliveryReceived: May 29, 2025Revised: July 17, 2025Published online: August 7, 2025[1] a) A. M. Vargason, A. C. Anselmo, S. Mitragotri, Nat. Biomed. Eng.2021, 5, 951; b) W. Graham, M. Torbett-Dougherty, A. Islam, S.Soleimani, T. A. Bruce-Tagoe, J. A. Johnson, Nanomaterials 2025, 15,285; c) T. Harimoto, W. H. Jung, D. J. Mooney, Nat. Rev. Mater. 2025,10, 191.[2] a) L. L.-W. Wang, Y. Gao, Z. Feng, D. J. Mooney, S. Mitragotri, Nat.Rev. Bioeng. 2024, 2, 944; b) Z. Zhao, A. Ukidve, J. Kim, S. Mitragotri,Cell 2020, 181, 151.[3] Q. T. H. Shubhra, Nat. Chem. 2024, 16, 1578.[4] a) M. J. Mitchell, M. M. Billingsley, R. M. Haley, M. E. Wechsler,N. A. Peppas, R. Langer, Nat. Rev. Drug Discovery 2021, 20, 101; b)A. Pusta, M. Tertis, I. Craciunescu, R. Turcu, S. Mirel, C. Cristea,Pharmaceutics 2023, 15, 1872; c) A. G. Almutary, P. Chauhan, L.Baldaniya, S. V. Menon, M. R. Kumar, B. Chaturvedi, N. Sharma, A.S. Chauhan, M. M. Abomughaid, M. Dhivyadharshni, A. C. Paiva-Santos, S. Lakhanpal, N. K. Jha, Nanomedicine 2025, 20, 5.[5] M. T. Manzari, Y. Shamay, H. Kiguchi, N. Rosen, M. Scaltriti, D. A.Heller, Nat. Rev. Mater. 2021, 6, 351.[6] a) K. Ariga, Bull. Chem. Soc. Jpn. 2024, 97, uoad001; b) W.Chaikittisilp, Y. Yamauchi, K. Ariga, Adv. Mater. 2022, 34,2107212.[7] K. Ariga, M. Nishikawa, T. Mori, J. Takeya, L. K. Shrestha, J. P. Hill, Sci.Technol. Adv. Mater. 2019, 20, 51.[8] A. Jancik-Prochazkova, K. Ariga, Research 2025, 8, 0624.[9] a) E. C. Yang, R. Divine, M. C. Miranda, A. J. Borst, W. Sheffler, J. Z.Zhang, J. Decarreau, A. Saragovi,M. Abedi, N. Goldbach,M. Ahlrichs,C. Dobbins, A. Hand, S. Cheng, M. Lamb, P. M. Levine, S. Chan, R.Skotheim, J. Fallas, G. Ueda, J. Lubner, M. Somiya, A. Khmelinskaia,N. P. King, D. Baker,Nat. Struct.Mol. Biol. 2024, 31, 1404; b)Q.Wang,R. Chang, X. Li, Y. Zhang, X. Fan, L. Shi, T. Li, Angew. Chem., Int. Ed.2025, 64, 202423004; c) S.-Y. Qin, J.-Q. Feng, Y.-J. Cheng, W.-L. Liu,A.-Q. Zhang, L. Wang, H. Wang, X.-Z. Zhang, Coord. Chem. Rev. 2024,502, 215600.Adv. Mater. 2025, 37, e10239 e10239 (16 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dewww.advancedsciencenews.com www.advmat.de[10] a) V. Karthik, S. Poornima, A. Vigneshwaran, D. P. R. D. D. Raj, R.Subbaiya, S. Manikandan, M. Saravanan, J. Mol. Struct. 2021, 1243,130844; b) J. W. Song, K. Kawakami, K. Ariga, Adv. Colloid InterfaceSci. 2025, 339, 103420.[11] a) C. Harkos, A. G. Hadjigeorgiou, C. Voutouri, A. S. Kumar, T.Stylianopoulos, R. K. Jain,Nat. Rev. Cancer 2025, 25, 324; b) C. Carini,A. A. Seyhan, J. Transl. Med. 2024, 22, 411.[12] a) I. H. Sarker, SNComput. Sci. 2022, 3, 158; b) B. Sanchez-Lengeling,A. Aspuru-Guzik, Science 2018, 361, 360; c) K. T. Butler, D. W. Davies,H. Cartwright, O. Isayev, A. Walsh, Nature 2018, 559, 547; d) C. J.Haug, J. M. Drazen, N. Engl. J. Med. 2023, 388, 1201.[13] a) Y. You, X. Lai, Y. Pan, H. Zheng, J. Vera, S. Liu, S. Deng, L. Zhang,Signal Transduction Targeted Ther. 2022, 7, 156; b) Y. Hong, B. Hou,H. Jiang, J. Zhang,Wires Comput. Mol. Sci. 2020, 10, 1450 c) T. H. Vo,N. T. K. Nguyen, Q. H. Kha, N. Q. K. Le, Comput. Struct. Biotechnol.J. 2022, 20, 2112.[14] a) Z. Li, P. Song, G. Li, Y. Han, X. Ren, L. Bai, J. Su, Mater. Today Bio2024, 25, 101014; b) S. Ebrahimian, M. K. Kalra, S. Agarwal, B. C.Bizzo, M. Elkholy, C. Wald, B. Allen, K. J. Dreyer, Acad. Radiol. 2022,29, 559.[15] a) L. Marques, B. Costa, M. Pereira, A. Silva, J. Santos, L. Saldanha,I. Silva, P. Magalhaes, S. Schmidt, N. Vale, Pharmaceutics 2024, 16,332; b) Z. Ahmed, K. Mohamed, S. Zeeshan, X. Dong,Database 2020,2020, baaa010.[16] K. Huanbutta, K. Burapapadh, P. Kraisit, P. Sriamornsak, T.Ganokratanaa, K. Suwanpitak, T. Sangnim, Eur. J. Pharm. Sci. 2024,203, 106938.[17] Z. Wang, J. Wang, J. Ayarza, T. Steeves, Z. Y. Hu, S. Manna, A. P. Esser-Kahn, Nat. Mater. 2021, 20, 869.[18] A. Aldoseri, K. N. Al-Khalifa, A. M. Hamouda, Appl. Sci. 2023, 13,7082.[19] a) S. Oh, J. Jekal, J. Liu, J. Kim, J.-U. Park, T. Lee, K.-I. Jang, Adv. Funct.Mater. 2024, 34, 2403562; b) O. M. D. Lemus, M. Cao, B. Cai, M.Cummings, D. Zheng, Cancers 2024, 16, 1206.[20] M. Mehta, T. A. Bui, X. Yang, Y. Aksoy, E. M. Goldys, W. Deng, ACSMater. Au 2023, 3, 600.[21] E. O. Pyzer-Knapp, J. W. Pitera, P. W. J. Staar, S. Takeda, T. Laino, D. P.Sanders, J. Sexton, J. R. Smith, A. Curioni, Npj Comput. Mater. 2022,8, 84.[22] Z. Nie, Y. Liu, L. Yang, S. Li, F. Pan, Adv. Energy Mater. 2021, 11,2003580.[23] a) A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek,S. Cholia, D. Gunter, D. Skinner, G. Ceder, K. A. Persson, APL Mater.2013, 1, 011002; b) A. S. Rosen, S. M. Iyer, D. Ray, Z. Yao, A. Aspuru-Guzik, L. Gagliardi, J. M. Notestein, R. Q. Snurr,Matter 2021, 4, 1578;c) S. K. Burley, R. Bhatt, C. Bhikadiya, C. Bi, A. Biester, P. Biswas, S.Bittrich, S. Blaumann, R. Brown, H. Chao, V. R. Chithari, P. A. Craig,G. V. Crichlow, J. M. Duarte, S. Dutta, Z. Feng, J. W. Flatt, S. Ghosh,D. S. Goodsell, R. K. Green, V. Guranovic, J. Henry, B. P. Hudson, M.Joy, J. T. Kaelber, I. Khokhriakov, J.-S. Lai, C. L. Lawson, Y. Liang, D.Myers-Turnbull, et al.,Nucleic Acids Res. 2024, 53, D564; d) D.Menon,D. Fairen-Jimenez,Matter 2025, 8, 101958.[24] S. Imani, X. Y. Li, K. Y. Chen, M. Maghsoudloo, P. J. Kaboli, M.Hashemi, S. Khoushab, X. P. Li, Front. Cell. Infect. Microbiol. 2025,14, 1501010.[25] a) I. H. Sarker, SN Comput. Sci. 2021, 2, 160.; b) T. Henser-Brownhill,L. Martin, P. Samangouei, A. Ladak, M. Apostolidou, B. Nagel, A.Kwok, Adv. Sci. 2024, 11, 2401935; c) A. Ortiz-Perez, D. van Tilborg, R.van der Meel, F. Grisoni, L. Albertazzi,Digital Discovery 2024, 3, 1280;d) M. Bhasin, G. P. Raghava, Protein Sci. 2004, 13, 596; e) S. Romero-Molina, Y. B. Ruiz-Blanco, J. Mieres-Perez, M. Harms, J. Munch, M.Ehrmann, E. Sanchez-Garcia, J. Proteome Res. 2022, 21, 1829; f) A.Tabet, T. Gebhart, G. Wu, C. Readman, M. P. Smela, V. K. Rana, C.Baker, H. Bulstrode, P. Anikeeva, D. H. Rowitch, O. A. Scherman,Phys. Chem. Chem. Phys. 2020, 22, 14976.[26] a) J. Schmidhuber, Neural Networks 2015, 61, 85; b) Z. Li, F. Liu,W. Yang, S. Peng, J. Zhou, IEEE Trans. Neural Networks Learn. Syst.2022, 33, 6999; c) X. Chen, H. Lv, NPG Asia Mater 2022, 14, 69; d) F.Mekki-Berrada, Z. Ren, T. Huang, W. K. Wong, F. Zheng, J. Xie, I. P. S.Tian, S. Jayavelu, Z. Mahfoud, D. Bash, K. Hippalgaonkar, S. Khan,T. Buonassisi, Q. Li, X. Wang, npj Comput. Mater. 2021, 7, 55; e) G.Fudenberg, D. R. Kelley, K. S. Pollard,Nat. Methods 2020, 17, 1111; f)M. A. K. Raiaan, S. Sakib, N. M. Fahad, A. A. Mamun, M. A. Rahman,S. Shatabda, M. S. H. Mukta, Decis. Anal. J. 2024, 11, 100470; g) X.Luo, Z. Wang, P. Gao, J. Lv, Y. Wang, C. Chen, Y. Ma, npj Comput.Mater. 2024, 10, 254; h) J. Noh, G. H. Gu, S. Kim, Y. Jung, Chem. Sci.2020, 11, 4871.[27] a) J. Qi, Q. Zhou, L. Lei, K. Zheng, Intell. Rob. 2021, 1, 18; b) V. S. S.Adury, A. Mukherjee, Digital Discovery 2024, 3, 705; c) W. Heyndrickx,L. Mervin, T. Morawietz, N. Sturm, L. Friedrich, A. Zalewski, A.Pentina, L. Humbeck, M. Oldenhof, R. Niwayama, P. Schmidtke, N.Fechner, J. Simm, A. Arany, N. Drizard, R. Jabal, A. Afanasyeva, R.Loeb, S. Verma, S. Harnqvist, M. Holmes, B. Pejo, M. Telenczuk,N. Holway, A. Dieckmann, N. Rieke, F. Zumsande, D. A. Clevert, M.Krug, C. Luscombe, et al., J. Chem. Inf. Model. 2024, 64, 2331.[28] a) S.Minaee, N. Kalchbrenner, E. Cambria, N. Nikzad,M. Chenaghlu,J. Gao, ACM Comput. Surv. 2022, 54, 62; b) X. Xiao, T. X. Trinh, Z.Gerelkhuu, E. Ha, T. H. Yoon, Comput. Struct. Biotechnol. J. 2024, 25,9.[29] S. M. Lundberg, S.-I. Lee, Proc. 31st Int. Conf. Neural Inf. Process. Syst.2017, 4768, https://doi.org/10.5555/3295222.3295230.[30] H. Jin, F. Chollet, Q. Song, X. Hu, J. Mach. Learn. Res. 2023, 24, 6.[31] a) J. Abramson, J. Adler, J. Dunger, R. Evans, T. Green, A. Pritzel, O.Ronneberger, L.Willmore, A. J. Ballard, J. Bambrick, S.W. Bodenstein,D. A. Evans, C.-C. Hung, M. O’Neill, D. Reiman, K. Tunyasuvunakool,Z. Wu, A. Zemgulyte, E. Arvaniti, C. Beattie, O. Bertolli, A. Bridgland,A. Cherepanov, M. Congreve, A. I. Cowen-Rivers, A. Cowie, M.Figurnov, F. B. Fuchs, H. Gladman, R. Jain, et al., Nature 2024, 630,493; b) J. L. Watson, D. Juergens, N. R. Bennett, B. L. Trippe, J. Yim,H. E. Eisenach, W. Ahern, A. J. Borst, R. J. Ragotte, L. F. Milles, B. I.M. Wicky, N. Hanikel, S. J. Pellock, A. Courbet, W. Sheffler, J. Wang,P. Venkatesh, I. Sappington, S. V. Torres, A. Lauko, V. De Bortoli, E.Mathieu, S. Ovchinnikov, R. Barzilay, T. S. Jaakkola, F. DiMaio, M.Baek, D. Baker, Nature 2023, 620, 1089.[32] a) B. Loef, A. Wong, N. A. H. Janssen, M. Strak, J. Hoekstra, H. S.J. Picavet, H. C. H. Boshuizen, W. M. M. Verschuren, G. M. Herber,Sci. Rep. 2022, 12, 10372; b) C. Cortes, V. Vapnik,Mach. Learn. 1995,20, 273; c) R. Rodríguez-Pérez, J. Bajorath, J. Comput.-Aided Mol. Des.2022, 36, 355.[33] a) M. Diqi, S. H. Mulyani, R. Pradila, SN Comput. Sci. 2023, 4, 396; b)G. Hinton, L. Deng, D. Yu, G. Dahl, A. Mohamed, N. Jaitly, A. Senior,V. Vanhoucke, P. Nguyen, T. Sainath, B. Kingsbury, IEEE Signal Process.Mag. 2012, 29, 82; c) Y. LeCun, B. Boser, J. S. Denker, D. Henderson,R. E. Howard, W. Hubbard, L. D. Jackel,Neural Comput. 1989, 1, 541;d) I. D. Mienye, T. G. Swart, G. Obaido, Information 2024, 15, 517.[34] a) F. Scarselli, M. Gori, A. C. Tsoi, M. Hagenbuchner, G. Monfardini,IEEE Trans. Neural Networks 2009, 20, 61; b) W. Zhung, H. Kim, W.Y. Kim, Nat. Commun. 2024, 15, 2688; c) J. Jumper, R. Evans, A.Pritzel, T. Green, M. Figurnov, O. Ronneberger, K. Tunyasuvunakool,R. Bates, A. Žídek, A. Potapenko, A. Bridgland, C. Meyer, S. A. A.Kohl, A. J. Ballard, A. Cowie, B. Romera-Paredes, S. Nikolov, R. Jain,J. Adler, T. Back, S. Petersen, D. Reiman, E. Clancy, M. Zielinski, M.Steinegger, M. Pacholska, T. Berghammer, S. Bodenstein, D. Silver,O. Vinyals, et al., Nature 2021, 596, 583; d) V. Valls, M. Zayats, A.Pascale, IEEE Int. Conf. Digital Health (ICDH), IEEE, Chicago, IL, USA,2023.Adv. Mater. 2025, 37, e10239 e10239 (17 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dehttps://doi.org/10.5555/3295222.3295230www.advancedsciencenews.com www.advmat.de[35] M. T. Ribeiro, S. Singh, C. Guestrin, Proc. 22nd ACM SIGKDD Int.Conf. Knowl. Discovery DataMin 2016, 1135, https://doi.org/10.1145/2939672.[36] B. J. Stear, T. M. Ahooyi, J. A. Simmons, C. Kollar, L. Hartman, K.Beigel, A. Lahiri, S. Vasisht, T. J. Callahan, C. M. Nemarich, J. C.Silverstein, D. M. Taylor, Sci. Data 2024, 11, 1338.[37] M. R. Kibria, R. I. Akbar, P. Nidadavolu, O. Havryliuk, S. Lafond, S.Azimi, Sci. Rep. 2023, 13, 547.[38] N. Iwano, T. Adachi, K. Aoki, Y. Nakamura, M. Hamada,Nat. Comput.Sci. 2022, 2, 378.[39] S. L. Lisanza, J. M. Gershon, S. W. K. Tipps, J. N. Sims, L. Arnoldt, S.J. Hendel, M. K. Simma, G. Liu, M. Yase, H. Wu, C. D. Tharp, X. Li,A. Kang, E. Brackenbrough, A. K. Bera, S. Gerben, B. J. Wittmann, A.C.McShan, D. Baker,Nat. Biotechnol. 2024, https://doi.org/10.1038/s41587-024-02456-0.[40] V. L. Deringer, A. P. Bartok, N. Bernstein, D. M. Wilkins, M. Ceriotti,G. Csanyi, Chem. Rev. 2021, 121, 10073.[41] C. Konig, A. Vellido, BioData Min. 2024, 17, 25.[42] W. E. Djeddi, K. Hermi, S. Ben Yahia, G. Diallo, BMC Bioinformatics2023, 24, 488.[43] a) K. Mi, W.-C. Chou, Q. Chen, L. Yuan, V. N. Kamineni, Y.Kuchimanchi, C. He, N. A. Monteiro-Riviere, J. E. Riviere, Z. Lin,J. Controlled Release 2024, 374, 219; b) R. T. Stiepel, E. S. Pena, S.A. Ehrenzeller, M. D. Gallovic, L. M. Lifshits, C. J. Genito, E. M.Bachelder, K. M. Ainslie, J. Controlled Release 2022, 351, 883; c) Z. Lin,W.-C. Chou, Y.-H. Cheng, C. He, N. A. Monteiro-Riviere, J. E. Riviere,Int. J. Nanomed. 2022, 17, 1365.[44] a) T. X. Trinh, M. Seo, T. H. Yoon, J. Kim, NanoImpact 2022, 25,100383; b) W.-C. Chou, Z. Lin, Toxicol. Sci. 2023, 191, 1; c) B. Singh,M. Crasto, K. Ravi, S. Singh, Intell. Pharm. 2024, 2, 598.[45] a) E. Bindewald, B. A. Shapiro, RNA 2006, 12, 342; b) C. Liu, H. Yang,Y. Feng, C. Liu, F. Rui, Y. Cao, X. Hu, J. Xu, J. Fan, Q. Zhu, J. Li, J. Clin.Transl. Hepatol. 2022, 10, 600.[46] L. Sun, H. Liu, Y. Ye, Y. Lei, R. Islam, S. Tan, R. Tong, Y. B. Miao, L.Cai, Signal Transduction Targeted Ther. 2023, 8, 418.[47] a) R. Tenchov, R. Bird, A. E. Curtze, Q. Zhou, ACS Nano 2021, 15,16982; b) J. Zhai, B. Fan, S. H. Thang, C. J. Drummond, Molecules2021, 26, 3648; c) F. Wang, P. Li, H. C. Chu, P. K. Lo, Biosensors 2022,12, 93; d) J. Wei, Y. Sun, H. Wang, T. Zhu, L. Li, Y. Zhou, Q. Liu, Z.Dai, W. Li, T. Yang, B. Wang, C. Zhu, X. Shen, Q. Yao, G. Song, Y.Zhao, H. Pei, Sci. Adv. 2024, 10, ado9880; e) D. Bae, H. Bae, H.-S. Yu,B. Dorjsembe, Y. H. No, T. Kim, N. H. Kim, J.-W. Kim, J. Kim, B.-S.Lee, Y. J. Kim, S. Park, Z. H. Khaleel, D. H. Sa, E.-C. Lee, J. C. Lee, J.Ham, J.-C. Kim, Y. H. Kim, Adv. Healthcare Mater. 2024, 13, 2303480.[48] a) V. Foglizzo, S. Marchio, Cancer 2022, 14, 2473; b) M. E. Rad, C.Soylukan, P. K. Kulabhusan, B. N. Gunaydin, M. Yuce, ACS Appl.Mater. Interfaces 2023, 15, 55201.[49] a) S. Z. Alshawwa, A. A. Kassem, R. M. Farid, S. K. Mostafa, G. S.Labib, Pharmaceutics 2022, 14, 883; b) S. Fathi-Karkan, M. Qindeel,R. Arshad, Z. Moafian, E. Ghazy, A. Rahdar, S. Ghotekar, Inorg. Chem.Commun. 2024, 161, 112028; c) A. U. Mahmood, M. M. Ghelardini,J. B. Tracy, Y. G. Yingling, Chem. Mater. 2024, 36, 9330.[50] a) S. M. Kawish, S. Sharma, P. Gupta, F. J. Ahmad, M. Iqbal,F. M. Alshabrmi, M. K. Anwer, S. Fathi-karkan, A. Rahdar, M. A.Aboudzadeh, Part. Part. Syst. Charact. 2024, 41, 2400049; b) G.Sinani, M. E. Durgun, E. Cevher, Y. Özsoy, Pharmaceutics 2023, 15,2021.[51] a) P. Zhan, A. Peil, Q. Jiang, D. Wang, S. Mousavi, Q. Xiong, Q. Shen,Y. Shang, B. Ding, C. Lin, Y. Ke, N. Liu, Chem. Rev. 2023, 123, 3976; b)J. Snoj, W. Zhou, A. Ljubetič, R. Jerala, Curr. Opin. Biotechnol. 2025,92, 103256.[52] a) X. Li, X. Peng, M. Zoulikha, G. F. Boafo, K. T. Magar, Y. Ju, W.He, Signal Transduction Targeted Ther. 2024, 9, 1; b) M. Shirzad, A.Salahvarzi, S. Razzaq, M. J. Javid-Naderi, A. Rahdar, S. Fathi-Karkan,A. Ghadami, Z. Kharaba, L. F. R. Ferreira, Crit. Rev. Oncol. Hematol.2025, 208, 104653.[53] a) J. Wang, Y. Li, G. Nie, Nat. Rev. Mater. 2021, 6, 766; b) M. Li, Q.Guo, C. Zhong, Z. Zhang, Drug Delivery 2023, 30, 2288797.[54] a) D. P. Kingma, M. Welling, Found. Trends Mach. Learn. 2019, 12,307; b) I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, Y. Bengio, Commun. ACM 2020, 63, 139.[55] a) X. Lin, Y. Rivenson, N. T. Yardimci, M. Veli, Y. Luo, M. Jarrahi, A.Ozcan, Science 2018, 361, 1004; b) K. Li, S. Yao, Z. Zhang, B. Cao, C.M. Wilson, D. Kalos, P. F. Kuan, R. Zhu, X. Wang, Bioinformatics 2022,38, 1631.[56] B. I. M. Wicky, L. F. Milles, A. Courbet, R. J. Ragotte, J. Dauparas, E.Kinfu, S. Tipps, R. D. Kibler, M. Baek, F. DiMaio, X. Li, L. Carter, A.Kang, H. Nguyen, A. K. Bera, D. Baker, Science 2022, 378, 56.[57] S. Rankovic, K. D. Carr, J. Decarreau, R. Skotheim, R. D. Kibler, S.Ols, S. Lee, J.-H. Chun, M. R. Tooley, J. Dauparas, H. E. Eisenach, M.Glögl, C. Weidle, A. J. Borst, D. Baker, N. P. King, Nat. Mater. 2025.[58] Y. Xu, S. Ma, H. Cui, J. Chen, S. Xu, F. Gong, A. Golubovic, M. Zhou,K. C. Wang, A. Varley, R. X. Z. Lu, B. Wang, B. Li,Nat. Commun. 2024,15, 6305.[59] a) W. Wang, K. Chen, T. Jiang, Y. Wu, Z. Wu, H. Ying, H. Yu, J. Lu, J.Lin, D. Ouyang,Nat. Commun. 2024, 15, 10804; b) B. Li, R. S. Manan,S. Q. Liang, A. Gordon, A. Jiang, A. Varley, G. Gao, R. Langer, W. Xue,D. Anderson, Nat. Biotechnol. 2023, 41, 1410.[60] S. Moayedpour, J. Broadbent, S. Riahi, M. Bailey, H. V. Thu, D.Dobchev, A. Balsubramani, R. N. D. Santos, L. Kogler-Anele, A.Corrochano-Navarro, S. Li, F. U. Montoya, V. Agarwal, Z. Bar-Joseph,S. Jager, Bioinformatics 2024, 40, btae342.[61] O. Kapustina, P. Burmakina, N. Gubina, N. Serov, V. Vinogradov, Ar-tif. Intell. Chem. 2024, 2, 100072.[62] a) C. Manzoni, D. A. Kia, J. Vandrovcova, J. Hardy, N. W. Wood, P.A. Lewis, R. Ferrari, Briefings Bioinf. 2018, 19, 286; b) L. Sikkema, C.Ramírez-Suástegui, D. C. Strobl, T. E. Gillett, L. Zappia, E.Madissoon,N. S.Markov, L.-E. Zaragosi, Y. Ji,M. Ansari,M.-J. Arguel, L. Apperloo,M. Banchero, C. Bécavin, M. Berg, E. Chichelnitskiy, M.-I. Chung, A.Collin, A. C. A. Gay, J. Gote-Schniering, B. H. Kashani, K. Inecik, M.Jain, T. S. Kapellos, T. M. Kole, S. Leroy, C. H. Mayr, A. J. Oliver, M.von Papen, L. Peter, et al., Nat. Med. 2023, 29, 1563.[63] X. Xu, X. Hua, H. Mo, S. Hu, J. Song, Basic Res. Cardiol. 2023, 118, 7.[64] D. Kim, T. Kobayashi, B. Voisin, J.-H. Jo, K. Sakamoto, S.-P. Jin, M.Kelly, H. B. Pasieka, J. L. Naff, J. H. Meyerle, I. D. Ikpeama, G. A.Fahle, F. P. Davis, S. D. Rosenzweig, J. C. Alejo, S. Pittaluga, H. H.Kong, A. F. Freeman, K. Nagao, Nat. Med. 2020, 26, 236.[65] C. Crema, G. Attardi, D. Sartiano, A. Redolfi, Front. Psychiatry 2022,13, 946387.[66] P. Cheng, Y. Peng, X.-L. Zhang, S. Chen, B.-B. Fang, Y.-Z. Li, Y.-M. Sun,Sci. Rep. 2024, 14, 21183.[67] J. M. Sagendorf, R. Mitra, J. Huang, X. S. Chen, R. Rohs, Biophys. Rev.2024, 16, 297.[68] P. Gainza, S. Wehrle, A. Van Hall-Beauvais, A. Marchand, A. Scheck,Z. Harteveld, S. Buckley, D. Ni, S. Tan, F. Sverrisson, C. Goverde, P.Turelli, C. Raclot, A. Teslenko, M. Pacesa, S. Rosset, S. Georgeon, J.Marsden, A. Petruzzella, K. Liu, Z. Xu, Y. Chai, P. Han, G. F. Gao, E.Oricchio, B. Fierz, D. Trono, H. Stahlberg, M. Bronstein, B. E. Correia,Nature 2023, 617, 176.[69] a) M. Azizi, R. Jahanban-Esfahlan, H. Samadian, M. Hamidi, K. Seidi,A. Dolatshahi-Pirouz, A. A. Yazdi, A. Shavandi, S. Laurent, M. B. O.Hagh, N. Kasaiyan, H. A. Santos, M.-A. Shahbazi, Mater. Today Bio2023, 20, 100672; b) Y. Wang, Y. Mo, Y. Sun, J. Li, Y. An, N. Feng, Y.Liu, J. Nanobiotechnol. 2024, 22, 669; c) M. Ramachandran, Z. Ma, K.Lin, C. De Souza, Y. Li, Nanoscale Adv. 2022, 4, 4470.[70] T. Wang, C. Wu, Y. Hu, Y. Zhang, J. Ma, RSC Adv. 2023, 13, 16488.[71] M. Chehelgerdi, M. Chehelgerdi, O. Q. B. Allela, R. D. C. Pecho, N.Jayasankar, D. P. Rao, T. Thamaraikani, M. Vasanthan, P. Viktor, N.Adv. Mater. 2025, 37, e10239 e10239 (18 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dehttps://doi.org/10.1145/2939672https://doi.org/10.1145/2939672https://doi.org/10.1038/s41587-024-02456-0https://doi.org/10.1038/s41587-024-02456-0www.advancedsciencenews.com www.advmat.deLakshmaiya, M. J. Saadh, A. Amajd, M. A. Abo-Zaid, R. Y. Castillo-Acobo, A. H. Ismail, A. H. Amin, R. Akhavan-Sigari,Mol. Cancer 2023,22, 169.[72] S. V. Torres, M. B. Valle, S. P. Mackessy, S. K. Menzies, N. R. Casewell,S. Ahmadi, N. J. Burlet, E.Muratspahic, I. Sappington,M. D. Overath,E. Rivera-de-Torre, J. Ledergerber, A. H. Laustsen, K. Boddum, A. K.Bera, A. Kang, E. Brackenbrough, I. A. Cardoso, E. P. Crittenden, R. J.Edge, J. Decarreau, R. J. Ragotte, A. S. Pillai, M. Abedi, H. L. Han, S.R. Gerben, A. Murray, R. Skotheim, L. Stuart, L. Stewart, et al.,Nature2025, 639, 225.[73] a) M. Gai, J. Simon, I. Lieberwirth, V. Mailänder, S. Morsbach, K.Landfester, Polym. Chem. 2020, 11, 527; b) L. Woythe, P. Madhikar,N. Feiner-Gracia, C. Storm, L. Albertazzi, ACS Nano 2022, 16, 3785.[74] J. Wang, J. Min, S. A. Eghtesadi, R. S. Kane, A. Chilkoti, ACS Nano2020, 14, 372.[75] L. Woythe, N. B. Tito, L. Albertazzi, Adv. Drug Delivery Rev. 2021, 169,1.[76] a) S. Y. Fam, C. F. Chee, C. Y. Yong, K. L. Ho, A. R. Mariatulqabtiah,W. S. Tan, Nanomaterials 2020, 10, 787; b) J. S. Suk, Q. Xu, N. Kim, J.Hanes, L. M. Ensign, Adv. Drug Delivery Rev. 2016, 99, 28.[77] M. D.McSweeney, T. Wessler, L. S. L. Price, E. C. Ciociola, L. B. Herity,J. A. Piscitelli, W. C. Zamboni, M. G. Forest, Y. Cao, S. K. Lai, J. Con-trolled Release 2018, 284, 171.[78] V. Dewaker, V. K. Morya, Y. H. Kim, S. T. Park, H. S. Kim, Y. H. Koh,Biomarker Res. 2025, 13, 52.[79] K. Shibata, Y. Shimoyama, Y. Orita, Ind. Eng. Chem. Res. 2023, 62,19683.[80] a) F. Di Meo, G. Fabre, K. Berka, T. Ossman, B. Chantemargue, M.Paloncýová, P. Marquet, M. Otyepka, P. Trouillas, Pharmacol. Res.2016, 111, 471; b) H. T. Hsueh, R. T. Chou, U. Rai, W. Liyanage, Y.C. Kim, M. B. Appell, J. Pejavar, K. T. Leo, C. Davison, P. Kolodziejski,A. Mozzer, H. Kwon, M. Sista, N. M. Anders, A. Hemingway, S. V. K.Rompicharla, M. Edwards, I. Pitha, J. Hanes, M. P. Cummings, L. M.Ensign, Nat. Commun. 2023, 14, 2509.[81] D. Cao, M. Chen, R. Zhang, Z. Wang,M. Huang, J. Yu, X. Jiang, Z. Fan,W. Zhang, H. Zhou, X. Li, Z. Fu, S. Zhang, M. Zheng, Nat. Methods2025, 22, 310.[82] G. Lunt, N. Hashemi, S. Mahajan, T. Tang, Sci. Rep. 2025, 15,4377.[83] Y. Goldstein, O. T. Cohen, O. Wald, D. Bavli, T. Kaplan, O. Benny, Sci.Adv. 2024, 10, adj4370.[84] A. H. Khalbas, T. M. Albayati, N. S. Ali, I. K. Salih, S. Afr. J. Chem. Eng.2024, 50, 261.[85] L. Möckl, Front. Cell Dev. Biol. 2020, 8, 253.[86] P. Tsiros, N. Chimarios, D. Zouraris, A. Tsoumanis, H. Sarimveis, G.Melagraki, I. Lynch, A. Afantitis, RSC Sustainability 2025, 3, 1494.[87] a) Mansi, P. Khanna, S. Yadav, A. Singh, L. Khanna, J. Biomol. Struct.Dyn. 2024, 1; b) J. Mao, J. Akhtar, X. Zhang, L. Sun, S. Guan, X. Li, G.Chen, J. Liu, H.-N. Jeon, M. S. Kim, K. T. No, G. Wang, iScience 2021,24, 103052.[88] W.-C. Chou, Q. Chen, L. Yuan, Y.-H. Cheng, C. He, N. A. Monteiro-Riviere, J. E. Riviere, Z. Lin, J. Controlled Release 2023, 361, 53.[89] a) L. Sheneman, G. Stephanopoulos, A. E. Vasdekis, PLoS One 2021,16, 0249196; b) E. Sadrfaridpour, T. Razzaghi, I. Safro, Mach. Learn.2019, 108, 1879.[90] A. V. Ponce-Bobadilla, V. Schmitt, C. S. Maier, S. Mensing, S.Stodtmann, Clin. Transl. Sci. 2024, 17, 70056.[91] a) L. Keutzer, H. You, A. Farnoud, J. Nyberg, S. G. Wicha, G. Maher-Edwards, G. Vlasakakis, G. K. Moghaddam, E. M. Svensson, M. P.Menden, Pharmaceutics 2022, 14, 1530; b) K. Wu, X. Li, Z. Zhou, Y.Zhao, M. Su, Z. Cheng, X. Wu, Z. Huang, X. Jin, J. Li, M. Zhang, J. Liu,B. Liu, Front. Pharmacol. 2024, 15, 1330855.[92] a) B. Huang,M. Abedi, G. Ahn, B. Coventry, I. Sappington, C. Tang, R.Wang, T. Schlichthaerle, J. Z. Zhang, Y.Wang, I. Goreshnik, C.W. Chiu,A. Chazin-Gray, S. Chan, S. Gerben, A. Murray, S. Wang, J. O’Neill, L.Yi, R. Yeh, A. Misquith, A. Wolf, L. M. Tomasovic, D. I. Piraner, M. J.D. Gonzalez, N. R. Bennett, P. Venkatesh, M. Ahlrichs, C. Dobbins,W. Yang, et al.,Nature 2025, 638, 796; b) N. Desai, D. Rana, S. Salave,D. Benival, D. Khunt, B. G. Prajapati,Molecules 2024, 29, 3131.[93] E. R. Gold, R. Cook-Deegan, Science 2025, 388, 131.[94] N. Baracaldo, A. Anwar, M. Purcell, A. Rawat, M. Sinn, B. Altakrouri,D. Balta, M. Sellami, P. Kuhn, U. Schopp, M. Buchinger, arXiv 2022,12443v1, https://doi.org/10.48550/arXiv.2202.12443.[95] H. J. Warraich, T. Tazbaz, R. M. Califf, JAMA, J. Am. Med. Assoc. 2025,333, 241.[96] C. S. Ajmal, S. Yerram, V. Abishek, V. P. M. Nizam, G. Aglave, J. D.Patnam, R. S. Raghuvanshi, S. Srivastava, AAPS J. 2025, 27, 22.[97] M. Shirzad, M. Shaban, V. Mohammadzadeh, A. Rahdar, S. Fathi-karkan, Z. S. Hoseini, M. Najafi, Z. Kharaba, M. A. Aboudzadeh, Bio-nanoscience 2025, 15, 354.[98] M. Alavinejad, M. Shirzad, M. J. Javid-Naderi, A. Rahdar, S. Fathi-karkan, S. Pandey,Med. Oncol. 2025, 42, 134.Adv. Mater. 2025, 37, e10239 e10239 (19 of 19) © 2025 The Author(s). Advanced Materials published by Wiley-VCH GmbH 15214095, 2025, 42, Downloaded from https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202510239 by National Institute For, Wiley Online Library on [23/10/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons Licensehttp://www.advancedsciencenews.comhttp://www.advmat.dehttps://doi.org/10.48550/arXiv.2202.12443 Artificial Intelligence-Driven Nanoarchitectonics for Smart Targeted Drug Delivery 1. Introduction 2. AI-Driven Workflow for Targeted Drug Delivery Systems 3. AI-Guided Process for Nanocarrier and Nanomaterial Design 3.1. Library-Based Nanocarrier Formulation 3.2. Iterative ML Model Development 3.3. Optimization of Nanocarrier Delivery Performance 4. A Three-Step Strategy for AI-Driven Smart Nanoarchitectonics 4.1. Rational Target Identification through Bioinformatic Profiling 4.2. ML-Guided Surface Engineering for Precision Targeting 4.3. Predictive Modeling of Binding and Delivery Kinetics 5. Challenges and Future Outlook Acknowledgements Conflict of Interest Keywords