Prediction and Interpretability of Glass Transition Temperature of Homopolymers by Data-Augmented Graph Convolutional Neural Networks

Hu, Junyang; Li, Zean; Lin, Jiaping; Zhang, Liangshun

doi:10.1021/acsami.3c13698

Cited by 7 publications

(3 citation statements)

References 74 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Additionally, molecular graphs can also serve as descriptors for training graph networks. [245] On the other hand, bulk material descriptors capture large-scale characteristics, including chain length and structure, and chemical and physical properties. In addition, synthesis methods and processing conditions also influence the properties of the polymers to a certain extent.…”

Section: Property Predictionmentioning

confidence: 99%

Machine Learning in Soft Matter: From Simulations to Experiments

Zhang,

Gong,

Jiang

2024

Adv Funct Materials

View full text Add to dashboard Cite

Soft matter with diverse functionalities that are easily designable has fascinated tremendous research interests in the past several decades. Nevertheless, the inherent confluence of time and length scale ubiquitous in soft matter immensely complicates the elucidation of the structure–property relationship and thereby severely impedes the function exploration of soft materials. Recently, the emergent machine learning (ML) techniques open new paradigms in property prediction and molecular design of functional materials, due to their extraordinarily distinguished performance in the aspect of trend identity and pattern extraction from data, and objective optimization by accelerating the guided search in high‐dimensional spaces. This review exclusively focuses on the current state‐of‐the‐art progress in the development of ML techniques applied in the realms of soft matter, ranging from coarse‐grained simulations to theoretical prediction on the structural formation and macroscopic properties, as well as the optimization and algorithm‐aided design in experiments. Finally, an outlook on the challenges and opportunities for this rapidly evolving field is discussed.

show abstract

Section: Property Predictionmentioning

confidence: 99%

Machine Learning in Soft Matter: From Simulations to Experiments

Zhang,

Gong,

Jiang

2024

Adv Funct Materials

View full text Add to dashboard Cite

show abstract

“…The structural formula of a monomer or repeat unit can be directly used in the form of a chemical graph or a line notation. Graph-convolutional neural networks take a chemical graph as input to predict the T g of polymers. − With this representation, substructure contribution to the prediction value within the chemical structure can be highlighted as an interpretation of the output of the model . This local interpretation is useful to understand important substructures within an input chemical structure.…”

Section: Introductionmentioning

confidence: 99%

“…12−14 With this representation, substructure contribution to the prediction value within the chemical structure can be highlighted as an interpretation of the output of the model. 12 This local interpretation is useful to understand important substructures within an input chemical structure. notation of the chemical structure can be directly utilized in combination with neural networks.…”

Section: ■ Introductionmentioning

confidence: 99%

Designing Heat-Resistant and Moldable Polyester Resin by the Integration of Machine Learning Models with Expert Knowledge

Zhang,

Miyao,

Izumiya

et al. 2024

ACS Appl. Polym. Mater.

View full text Add to dashboard Cite

Polyester resin has advantages in transparency and chemical resistance and is widely used in films and containers. In industrial applications, multiple conflicting properties of polyester resin must be optimized. Nevertheless, few reports have been found dealing with the design of polyester resins with machine learning (ML) models. Herein, we report a multiobjective design strategy of heat tolerant and moldable polyester resin, represented by the glasstransition temperature (T g ) and the softening point (SP). Our proposed workflow is an interplay between ML models and expert knowledge. Highly accurate interpretable linear regression models using chemical structural features were constructed for T g and SP, which were utilized for evaluating previously uninvestigated monomers. Insights into substructures with which highly correlated properties (T g and SP) were compromised were obtained by analyzing the regression coefficients of a linear model for SP/T g . Based on the insight from the SP/T g model, four dicarboxylic monomers consisting of untested molecular scaffolds were proposed and with which polyester resins were actually synthesized. The synthesized resins exhibited desired properties, consistent with prediction results by ML models. The reported workflow successfully proposed the dicarboxylic monomers with which polyester resins had desirable multiple properties.

show abstract