Convergence of Artificial Intelligence, Machine Learning, Cheminformatics, and Polymer Science in Macromolecules
2024; American Chemical Society; Linguagem: Inglês
10.1021/acs.macromol.4c01704
ISSN1520-5835
AutoresArthi Jayaraman, Bradley D. Olsen,
Tópico(s)Computational Drug Discovery Methods
ResumoInfoMetricsFiguresRef. MacromoleculesVol 57/Issue 16Article This publication is free to access through this site. Learn More CiteCitationCitation and abstractCitation and referencesMore citation options ShareShare onFacebookX (Twitter)WeChatLinkedInRedditEmailJump toExpandCollapse EditorialAugust 9, 2024Convergence of Artificial Intelligence, Machine Learning, Cheminformatics, and Polymer Science in MacromoleculesClick to copy article linkArticle link copied!Arthi JayaramanArthi JayaramanMore by Arthi Jayaramanhttps://orcid.org/0000-0002-5295-4581Bradley OlsenBradley OlsenMore by Bradley Olsenhttps://orcid.org/0000-0002-7272-7140Open PDFMacromoleculesCite this: Macromolecules 2024, 57, 16, 7685–7688Click to copy citationCitation copied!https://pubs.acs.org/doi/10.1021/acs.macromol.4c01704https://doi.org/10.1021/acs.macromol.4c01704Published August 9, 2024 Publication History Received 19 July 2024Published online 9 August 2024Published in issue 27 August 2024editorialCopyright © 2024 American Chemical Society. This publication is available under these Terms of Use. Request reuse permissionsThis publication is licensed for personal use by The American Chemical Society. ACS PublicationsCopyright © 2024 American Chemical SocietySubjectswhat are subjectsArticle subjects are automatically applied from the ACS Subject Taxonomy and describe the scientific concepts and themes of the article.Chemical structureMolecular dynamics simulationsMolecular modelingPolymer sciencePolymersOver the past decade, it has become abundantly clear that there is no escaping the emerging growth of artificial intelligence (AI) in every facet of our life. The advanced algorithms and modeling approaches grouped under this umbrella "AI" term are empowering us to use data to build predictive models for complex phenomena and to apply these models in generative contexts to make suggestions or recommendations. For example, AI models can be used to predict the properties of small molecules, and generative models can be used to recommend synthetic routes to most effectively make those molecules by quantifying the likelihood of success of different reaction pathways; image-processing models can classify medical images to identify health or disease states, while generative models can create images from text prompts. Scientists and engineers appreciate and understand the myriad technological advancements that AI can enable and are working diligently to develop the necessary algorithms and software and to gather the required training data to harness the power of AI. Simultaneously, social scientists are grasping the ethical, legal, and societal implications of this exponential growth of AI in various aspects of our lives.As Associate Editors of Macromolecules and as active researchers in the area of AI and machine learning (ML) for polymer science, it is not lost on us that the field of polymer science is lagging behind most other areas of chemistry in the adoption and application of AI tools. Specifically, the fields of inorganic chemistry, nanomaterials, biomedical imaging, and bioinformatics are farther ahead in leveraging the power of AI and ML. We think that this may be because our field presents many unique problems that prevent the AI tools developed in other areas of chemistry from being directly translated to polymer science, some of which we address below.Nonetheless, through this editorial we want to share with our readers examples of how some researchers in the polymer community have successfully developed/extended and used AI and ML methods for accelerating polymer discovery and innovation. We describe briefly more than 30 articles published over the last 4 years in Macromolecules that we believe serve as exemplary studies highlighting AI and ML for enabling polymer design and discovery, for high-throughput polymer synthesis and characterization, complementing molecular modeling and simulations, and improving fundamental understanding of results from experiments and/or simulations.Making Polymer Data Widely Available and Easily UsableClick to copy section linkSection link copied!ML and many other data-driven tasks have as their foundation the availability of data. Before we can develop ML models, the first nontrivial and time-consuming task involves collection or generation of data. Data can come from experiments, theory, or molecular simulation. However, the simple existence of the data is not enough: the community must be able to find and use the data. The dearth of universal databases with information about polymers (e.g., properties and processing conditions) and consistent polymer representation in these databases are currently the biggest barriers to more prevalent use of ML modeling in polymer science.While human beings rely on chemical structure diagrams, computers require machine-readable representation of polymers (i.e., fingerprints, descriptors, line notations, or graphs) to perform important operations such as search, comparing the similarity of polymers, ML modeling, and automated analysis. In some studies, authors evaluated various forms of descriptors to identify the best representation for the problem at hand. For example, Ethier et al. described several types of mathematical descriptors (Morgan fingerprints, molecular descriptors, etc.) that could best encode polymer–solvent interactions. (1) Similarly, Kuenneth et al. also considered various forms of polymer fingerprinting for predicting glass transition, melting temperature, and degradation temperatures of homopolymers and copolymers. (2,3)With the goal of creating a universal machine-readable representation for polymers, Deagen et al. developed applications for BigSMILES, a machine-readable representation of polymer chemical structure that requires specialized grammar and syntax. (4) They also demonstrated the "algorithmic translation between chemical structure diagrams and BigSMILES line notation" to facilitate seamless interconversion of the machine-readable form to and from the polymer chemistry language. (4) Shi et al. also discussed the idea of similarity between polymers which is required if one wishes to use ML models to perform the tasks of ranking, clustering, and classification during a design optimization or characterization analysis problem. (5) The implementation of computer science techniques combined with polymer researchers' domain-specific expertise opens doors to unlimited collaborations, although consensus must be found on how we format, host, and share data.Accelerating ExperimentationClick to copy section linkSection link copied!There is a major push broadly in chemical and materials sciences to create automated or autonomous experiments to accelerate the pace with which we can learn key structure–property relationships (exploration) or develop a material with a target function or property (exploitation). This requires hardware for high-throughput (HT) experimentation and software for experimental design and decision-making that optimally allocates experimental resources to achieve the objective of the experimental campaign. While automation has existed for some time, decreasing hardware costs and technological advances have made integration ever more possible. For example, Vittoria et al. developed a HT workflow covering the entire cycle from polymer synthesis, structural, mechanical, and rheological characterization to make olefin block copolymers. (6) Similarly, Urciuoli et al. used HT experimentation to synthesize ethylene/hex-1-ene statistical multiblock copolymers over a range of compositions through chain shuttling copolymerization. (7)Polymer scientists are also integrating HT experiments with rapid analysis; Richbourg et al. used HT fluorescence recovery after photobleaching (FRAP) experiments and characterized the diffusion coefficients of small and large molecules through numerous hydrogel formulations. (8) Similarly, Meleties et al. demonstrated the use of HT microrheological assay to assess the gelation kinetics of a coiled-coil protein across a compositional space with varying ionic strengths and pH values. (9) One of the greatest advantages of ML is the ability to grapple with huge data sets that result from such HT experiments and analyses; Yin et al. exemplified this by building an in situ HT investigation system using synchrotron X-ray to obtain fast millisecond-resolved structural evolution of semicrystalline polymers. (10) They claim that "various structural types of information including crystallization kinetics, polymorphism, and the growth and orientation of lamellar crystals under an actual environment of non-isothermal crystallization with a cooling gradient, intense flow, and high-pressure during injection molding have been revealed for the first time." Data wrangling that has previously taken months to years is now being completed in minutes to hours.Learning from Our Data FasterClick to copy section linkSection link copied!With high-throughput experimentation comes the next challenge of needing ML workflows for fast (and objective) analysis and interpretation of the large amounts of data generated from these experiments. Polymer researchers can borrow methods/approaches well-established in computer vision and pattern recognition to analyze patterns in the 2D images, such as polymer characterization data that are collected as two-dimensional (2D) images (e.g., microscopy and scattering data). Qu et al. demonstrated this when they described a deep-learning-based image analysis method to quantify the distribution of spherical particles in a polymer matrix from transmission electron microscopy (TEM) images. (11)ML and optimization tools can also be valuable in interpretation of other forms of characterization data (e.g., intensity vs wavevector, absorbance vs wavelength, or molar mass vs elution time). For example, to interpret small-angle scattering (SAS) profiles, ML-based Computational Reverse Engineering Analysis of Scattering Experiments (ML-CREASE) introduced by Wu and Jayaraman uses a combination of ML models and genetic algorithms to identify structures (or structural features) whose computed scattering profile most closely match experimentally measured scattering profiles. (12)In cases where one form of characterization data is hard to interpret, one can use ML to link that data to easier-to-interpret data from other measurement(s). To address the challenge that near-infrared (NIR) spectroscopy "does not provide insights into the chain composition, conformation, and topology of polyolefins", Sutliff et al. used ML models to correlate structural information from slower measurement methods to NIR spectra. (13) Similarly, using ML models to connect information from one type of measurement to another type of information, Rajabifar et al. trained a data-driven approach to output a surface's viscoelastic and adhesive properties from the surface's experimentally acquired atomic force microscopy (AFM) data. (14) When AI tools take on the cognitively expensive data wrangling and understanding tasks, researchers become free to think creatively toward new avenues of investigation.ML-Driven Design and DiscoveryClick to copy section linkSection link copied!Most often, researchers work toward optimizing polymer design (e.g., composition, architecture, molecular weights, and dispersity index) to achieve target properties (e.g., compatibilization, mechanical properties, and glass transition temperatures). ML is empowering researchers in this quest by predicting which polymers will have optimal performance and properties without the need to experimentally explore large design spaces, thereby accelerating the process of design and discovery. Important examples include the prediction of glass transition temperatures in nanocomposites by the NanoMine team led by Ma et al. or similar predictions for novel polymer structures by Miccio et al., (15,16) predicting the compatibilization efficiency of graft copolymers by Zhou et al., (17) ML for the selection of synthesis conditions by Park et al., or the PolyID tool for the design of new biomass-based polymers by Wilson et al. at the National Renewable Energy Lab (NREL). (18,19)ML can overcome materials design challenges with multiobjective optimization. Wheatle et al. showed this for the simultaneous tuning of ionic conductivity and mechanical polymers in polymer electrolytes. (20) They used molecular dynamics (MD) simulations to generate the corresponding simulated measurements for varying design parameters, namely, polymer molar mass and polarity, and using Bayesian optimization identified the Pareto front that maximizes ion transport and electrolyte mechanical properties.ML modeling can also aid in identifying processing conditions or additives that help to achieve targeted performance. For example, Aoki et al. developed an ML workflow to predict Flory–Huggins χ parameters for polymer–solvent pairs using a hybrid data set including experimentally-observed and quantum-chemically calculated χ parameters for polymer–solvent pairs. (21) Along similar lines, Ethier et al. utilized curated data for cloud point temperatures for polymer–solvent combinations to construct entire phase diagrams of polymer solutions using ML predictions. (1) Chandrasekaran et al. also developed a deep neural network (DNN) trained on a larger data set of over 4500 polymers and their corresponding solvents/nonsolvents that classifies solvents and nonsolvents into clusters identified as nonpolar, polar–aprotic, and polar–protic behavior. (22) We expect one could extend these approaches for identifying additives that, when added to the key polymer ingredient, retains the attractive aspects of the polymer but improves the product function (e.g., rheological modifiers in paints) or processability (e.g., by lowering glass transition).Combining Machine Learning and Molecular Modeling and SimulationsClick to copy section linkSection link copied!ML can also be used to accelerate molecular simulations, especially in cases where the specific insights one wants could only be obtained by running computationally expensive, long time scale simulations.Schneider and de Pablo trained an ML model to quantify late-time dynamics of block copolymer relaxation from coarse-grained simulations capturing defect kinetics in the lamellar copolymer morphology as it approaches equilibrium. (23) Yong and Kim used DNNs to accelerate Langevin field-theoretic simulations (L-FTS) that otherwise take a long time to predict ensemble averages of thermodynamic quantities. (24) Kwon et al. used ML to a similar end, reducing the computational time needed to calculate ion transport properties from MD simulations. (25) Interestingly, they also found that descriptors containing information about ion clustering and time evolution of ion transport properties are more advantageous for their goal, compared with polymer structure-based descriptors that only contain information about the polymer and not the ion–polymer interactions. This connects back to our discussion of finding appropriate machine-readable representation of each system.ML can be used to analyze the trajectories from molecular simulations when one wishes to investigate mechanisms that traditional simulation analysis may fail to provide. For example, Zhou et al. used ML methods to investigate the microscopic mechanisms of self-healing in polymers in MD simulations. (26) Ziolek et al. used unsupervised ML methods to gain insight into the polymer conformations within the core–shell micelles of pluronics. (27) Bhattacharya et al. developed a DNN to predict coil–globule phase transitions in polymers, and Jin et al. described a decision tree to investigate the adsorption of biomimetic random heteropolymers on polar, nonpolar, and charged groups onto graphene oxide and carbon nanotube. (28,29) Bera et al. combined all-atom MD simulations and unsupervised ML techniques to study the water–water hydrogen bonds inside the anionic poly(acrylic acid) brushes. (30) Conversely, Braghetto et al. used supervised ML, in the form of the long–short-term memory neural network, to assist in the distinction between polymer knots confined within a spherical cavity. (31)We hope that these articles in Macromolecules serve as excellent examples that inspire the polymer science community to adopt such approaches as core pillars of data stewardship, experimental design, data analysis, and material discovery to accelerate their own research.Lastly, as we put together this editorial, we could not help but notice the lack of diversity in gender identity among the active researchers in ML for polymers, an issue that needs to be addressed. The polymer community has to redouble efforts, both as individuals and as organizations, to address and help remedy this situation. We can achieve change by being more thoughtful and deliberate in our lab practices to create more inclusive work environments, by paying closer attention to the diversity and order of speakers when organizing conferences, and by being more deliberate in seeking outstanding researchers with different gender identities for speaking and leadership positions so that our younger researchers see them as role models for our field.We at Macromolecules are excited to see increasing submissions of manuscripts that highlight innovative and impactful developments and use of AI and ML in polymers that push the frontiers of fundamental polymer science forward.Author InformationClick to copy section linkSection link copied!Corresponding AuthorArthi Jayaraman, https://orcid.org/0000-0002-5295-4581AuthorBradley Olsen, https://orcid.org/0000-0002-7272-7140NotesViews expressed in this editorial are those of the authors and not necessarily the views of the ACS.ReferencesClick to copy section linkSection link copied! This article references 31 other publications. 1Ethier, J. G.; Casukhela, R. K.; Latimer, J. J.; Jacobsen, M. D.; Rasin, B.; Gupta, M. K.; Baldwin, L. A.; Vaia, R. A. Predicting Phase Behavior of Linear Polymers in Solution Using Machine Learning. Macromolecules 2022, 55 (7), 2691– 2702, DOI: 10.1021/acs.macromol.2c00245 Google Scholar5Predicting Phase Behavior of Linear Polymers in Solution Using Machine LearningEthier, Jeffrey G.; Casukhela, Rohan K.; Latimer, Joshua J.; Jacobsen, Matthew D.; Rasin, Boris; Gupta, Maneesh K.; Baldwin, Luke A.; Vaia, Richard A.Macromolecules (Washington, DC, United States) (2022), 55 (7), 2691-2702CODEN: MAMOBX; ISSN:0024-9297. (American Chemical Society) The phase behavior of polymers in soln. is crucial to many applications in polymer processing, synthesis, self-assembly, and purifn. Quant. prediction of polymer soly. space for an arbitrary polymer-solvent pair and across a large compn. range is challenging. Qual. agreement is provided by many current theor. models, but only a portion of the phase space is quant. predicted. Here, we utilize a curated database for binary polymer solns. comprised of 21 linear polymers, 61 solvents, and 97 unique polymer-solvent combinations (6524 cloud point temps.) to construct phase diagrams from machine learning predictions. A generalizable feature vector is developed that includes component descriptors concatenated with state variables and an exptl. data descriptor (phase direction). The impact of several types of descriptors (Morgan fingerprints, mol. descriptors, Hansen soly. parameters) to encode polymer-solvent interactions is assessed. Hansen soly. parameters (HSPs) are also introduced as a means to understand the general breadth of linear polymer-solvent space, as well as d. and distribution of curated data. Two common regression algorithms (XGBoost and neural networks) establish the generality of the descriptors; provide a root mean squared error (RMSE) within 3°C for predicted cloud points in the test set; and offer excellent agreement with upper and lower crit. soly. curves, isopleths, and closed-loop phase behavior by a single model. The ability to extrapolate to polymers that are very dissimilar from the curated data is poor, but with as little as 20 cloud points or a single phase boundary, RMSE error of predictions are within 5°C. This implies that the current model captures aspects of the underlying physics and can readily exploit correlations to reduce required data for addnl. polymer-solvent pairs. Finally, the model and data are accessible via the Polymer Property Predictor and Database (3PDb). >> More from SciFinder ®https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB38XnvVKhsrk%253D&md5=c77c01e0ff4b3922f0fb3656a9b4bed52Kuenneth, C.; Schertzer, W.; Ramprasad, R. Copolymer Informatics with Multitask Deep Neural Networks. Macromolecules 2021, 54 (13), 5957– 5961, DOI: 10.1021/acs.macromol.1c00728 Google Scholar6Copolymer Informatics with Multitask Deep Neural NetworksKuenneth, Christopher; Schertzer, William; Ramprasad, RampiMacromolecules (Washington, DC, United States) (2021), 54 (13), 5957-5961CODEN: MAMOBX; ISSN:0024-9297. (American Chemical Society) Polymer informatics tools have been recently gaining ground to efficiently and effectively develop, design, and discover new polymers that meet specific application needs. So far, however, these data-driven efforts have largely focused on homopolymers. Here, we address the property prediction challenge for copolymers, extending the polymer informatics framework beyond homopolymers. Advanced polymer fingerprinting and deep-learning schemes that incorporate multitask learning and meta learning are proposed. A large data set contg. over 18 000 data points of glass transition, melting, and degrdn. temp. of homopolymers and copolymers of up to two monomers is used to demonstrate the copolymer prediction efficacy. The developed models are accurate, fast, flexible, and scalable to more copolymer properties when suitable data become available. >> More from SciFinder ®https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB3MXhsVWrt7nP&md5=0ad5fd62467c2ac72e38f83e8d4beec83Kuenneth, C.; Schertzer, W.; Ramprasad, R. Correction to "Copolymer Informatics with Multitask Deep Neural Networks. Macromolecules 2021, 54 (15), 7321– 7321, DOI: 10.1021/acs.macromol.1c01539 Google ScholarThere is no corresponding record for this reference.4Deagen, M. E.; Dalle-Cort, B.; Rebello, N. J.; Lin, T.-S.; Walsh, D. J.; Olsen, B. D. Machine Translation between BigSMILES Line Notation and Chemical Structure Diagrams. Macromolecules 2024, 57 (1), 42– 53, DOI: 10.1021/acs.macromol.3c01378 Google ScholarThere is no corresponding record for this reference.5Shi, J.; Rebello, N. J.; Walsh, D.; Zou, W.; Deagen, M. E.; Leao, B. S.; Audus, D. J.; Olsen, B. D. Quantifying Pairwise Similarity for Complex Polymers. Macromolecules 2023, 56 (18), 7344– 7357, DOI: 10.1021/acs.macromol.3c00761 Google Scholar9Quantifying Pairwise Similarity for Complex PolymersShi, Jiale; Rebello, Nathan J.; Walsh, Dylan; Zou, Weizhong; Deagen, Michael E.; Leao, Bruno Salomao; Audus, Debra J.; Olsen, Bradley D.Macromolecules (Washington, DC, United States) (2023), 56 (18), 7344-7357CODEN: MAMOBX; ISSN:0024-9297. (American Chemical Society) Defining the similarity between chem. entities is an essential task in polymer informatics, enabling ranking, clustering, and classification. Despite its importance, the pairwise chem. similarity for polymers remains an open problem. Here, a similarity function for polymers with well-defined backbones is designed based on polymers' stochastic graph representations generated from canonical BigSMILES, a structurally based line notation for describing macromols. The stochastic graph representations are sepd. into three parts: repeat units, end groups, and polymer topol. The earth mover's distance is utilized to calc. the similarity of the repeat units and end groups, while the graph editing distance is used to calc. the similarity of the topol. These three values can be linearly or nonlinearly combined to yield an overall pairwise chem. similarity score for polymers that is largely consistent with the chem. intuition of expert users and is adjustable based on the relative importance of different chem. features for a given similarity problem. This method gives a reliable soln. to quant. calc. the pairwise chem. similarity score for polymers and represents a vital step toward building search engines and quant. design tools for polymer data. >> More from SciFinder ®https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB3sXhvVejtLjM&md5=059fec3cd24116655e310005bb1aff0e6Vittoria, A.; Urciuoli, G.; Costanzo, S.; Tammaro, D.; Cannavacciuolo, F. D.; Pasquino, R.; Cipullo, R.; Auriemma, F.; Grizzuti, N.; Maffettone, P. L.; Busico, V. Extending the High-Throughput Experimentation (HTE) Approach to Catalytic Olefin Polymerizations: From Catalysts to Materials. Macromolecules 2022, 55 (12), 5017– 5026, DOI: 10.1021/acs.macromol.2c00813 Google Scholar10Extending the High-Throughput Experimentation (HTE) Approach to Catalytic Olefin Polymerizations: From Catalysts to MaterialsVittoria, Antonio; Urciuoli, Gaia; Costanzo, Salvatore; Tammaro, Daniele; Cannavacciuolo, Felicia Daniela; Pasquino, Rossana; Cipullo, Roberta; Auriemma, Finizia; Grizzuti, Nino; Maffettone, Pier Luca; Busico, VincenzoMacromolecules (Washington, DC, United States) (2022), 55 (12), 5017-5026CODEN: MAMOBX; ISSN:0024-9297. (American Chemical Society) In this study, a state-of-the-art high-throughput experimentation (HTE) workflow for catalytic olefin polymn., covering an unprecedented wide part of the polymer knowledge and value chains from catalytic synthesis all the way down to "engineering" microrheol., was thoroughly assessed with respect to its ability to prep. new materials and produce large and accurate databases for the investigation of quant. structure-property relationships (QSPRs). Olefin blocks copolymers (OBCs) produced under chain-shuttling polymn. conditions were used as a demonstration case. The results of a thorough microstructural, structural, mech., morphol., and rheol. characterization of OBC replicas prepd. with the HTE synthetic platform and a com. sample, chosen as a benchmark, demonstrate the robustness of the approach. The proposed workflow can become a paradigm for the high-throughput synthesis and investigation of novel materials, thus reducing the time to market of new products. In our opinion, this opens the door to integrated HTE and artificial intelligence approaches to QSPR problem solving in the numerous cases for which a thorough understanding of the theory is not sufficient to deterministically unravel the complexity of practical applications. >> More from SciFinder ®https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB38XhsFWnsLrL&md5=62becf121dbfd01aba26cecef802e09d7Urciuoli, G.; Vittoria, A.; Cannavacciuolo, F. D.; Cipullo, R.; Costanzo, S.; Ianniello, V.; Bellavista, F.; Ruiz de Ballesteros, O.; Busico, V.; Grizzuti, N.; Auriemma, F. Effect of Segregation Strength on Mesophase Separation in Statistical Multiblock Copolymers Synthesized through a High-Throughput Experimentation Approach. Macromolecules 2023, 56 (24), 10163– 10178, DOI: 10.1021/acs.macromol.3c01958 Google ScholarThere is no corresponding record for this reference.8Richbourg, N. R.; Peppas, N. A. High-Throughput FRAP Analysis of Solute Diffusion in Hydrogels. Macromolecules 2021, 54 (22), 10477– 10486, DOI: 10.1021/acs.macromol.1c01752 Google Scholar12High-Throughput FRAP Analysis of Solute Diffusion in HydrogelsRichbourg, Nathan R.; Peppas, Nicholas A.Macromolecules (Washington, DC, United States) (2021), 54 (22), 10477-10486CODEN: MAMOBX; ISSN:0024-9297. (American Chemical Society) Increasingly accurate math. models have been developed to relate solute and hydrogel properties to solute diffusion coeffs. in hydrogels, primarily by comparing solute sizes and hydrogel mesh sizes. Here, we use a standardized, high-throughput method for fluorescence recovery after photobleaching (FRAP) expts. and anal. to characterize the diffusion coeffs. of fluorescein, three sizes of FITC-dextran, and three sizes of FITC-conjugated poly(ethylene glycol) (PEG) through 18 structurally varied poly(vinyl alc.) (PVA) hydrogel formulations. Increasing the hydrogel mesh radii increased the diffusivities of all the tested solutes within the hydrogels. While the diffusivity of FITC-dextrans in hydrogels decreased with increasing solute size, the diffusivity of FITC-PEGs increased with increasing solute size, suggesting that a generalized hydrodynamic radius-based model is not universally applicable for solute diffusion in hydrogels. The high-throughput characterization method for solute diffusion in hydrogels described here facilitates precise hydrogel design for biomedical applications. >> More from SciFinder ®https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BB3MXisVegu7rM&md5=46e647b37255142e542
Referência(s)