A HYBRID APPROACH FOR PREDICTING GLASS TRANSITION TEMPERATURES OF ORGANIC HOMOPOLYMERS: QSPR MODELING COMBINED WITH THE INCREMENTAL METHOD
Rubrics: 2. CHEMICAL TECHNOLOGY
Authors:
1.
Kazan National Research Technological University
2.
Kazan National Research Technological University
3.
Kazan National Research Technological University
4.
Kazan National Research Technological University
5.
KNITU
6.
Kazan National Research Technological University
(Department of General Chemical Technology, Assistant)
employee from 01.01.2024 to 01.01.2025
Kazan, Kazan, Russian Federation
employee from 01.01.2024 to 01.01.2025
Kazan, Kazan, Russian Federation
7.
KNITU
8.
KNITU
Type:
Article
Pages:
from 68
to 74
Status:
In work
Language:
Russian
Keywords:
MACHINE LEARNING ALGORITHMS, INCREMENTAL METHOD, QSPR MODEL, ORGANIC HOMOPOLYMERS
Abstract (English):
Accurate prediction of the glass transition temperature, one of the most critical characteristics of polymers, is key to developing and applying polymers with desired properties. Traditionally, polymer glass transition temperature prediction has relied on semi-empirical methods, such as the A.A. Askadskii incremental method; however, the development of computational technologies and machine learning algorithms is opening new avenues for improving prediction accuracy. The aim of this work was to create a hybrid approach for predicting the glass transition temperature of organic homopolymers by combining the A.A. Askadskii method and a QSPR model. This combines the advantages of theoretical analysis of homopolymer glass transition temperatures with the capabilities of machine learning for more accurate polymer property prediction. The following machine learning algorithms were used in this study: Random Forest, K-Nearest Neighbors, and Multilayer Perceptron (MLP). The molecular structure of the polymers was represented using two types of descriptors: MACCSKeys and Morgan fingerprints, which reflect different aspects of the structure of the repeating units of organic homopolymers. To improve prediction accuracy, the hyperparameters of the Random Forest algorithm were optimized, achieving a coefficient of determination (R2) of up to 0.77 on the test set. A comparative analysis of the effectiveness of various machine learning algorithms and descriptor types was performed. It was found that the use of Morgan fingerprints, which account for the spatial arrangement of molecular fragments, provides higher prediction accuracy compared to structural keys, which only reflect the presence or absence of certain structural elements. Special attention was paid to predicting the glass transition temperatures of isomeric organic homopolymers, for which accounting for the spatial arrangement of substituents is critical. The results demonstrate the promise of using machine learning methods for predicting polymer glass transition temperatures based on glass transition theories, and indicate the need for further research into similar hybrid models.
Accurate prediction of the glass transition temperature, one of the most critical characteristics of polymers, is key to developing and applying polymers with desired properties. Traditionally, polymer glass transition temperature prediction has relied on semi-empirical methods, such as the A.A. Askadskii incremental method; however, the development of computational technologies and machine learning algorithms is opening new avenues for improving prediction accuracy. The aim of this work was to create a hybrid approach for predicting the glass transition temperature of organic homopolymers by combining the A.A. Askadskii method and a QSPR model. This combines the advantages of theoretical analysis of homopolymer glass transition temperatures with the capabilities of machine learning for more accurate polymer property prediction. The following machine learning algorithms were used in this study: Random Forest, K-Nearest Neighbors, and Multilayer Perceptron (MLP). The molecular structure of the polymers was represented using two types of descriptors: MACCSKeys and Morgan fingerprints, which reflect different aspects of the structure of the repeating units of organic homopolymers. To improve prediction accuracy, the hyperparameters of the Random Forest algorithm were optimized, achieving a coefficient of determination (R2) of up to 0.77 on the test set. A comparative analysis of the effectiveness of various machine learning algorithms and descriptor types was performed. It was found that the use of Morgan fingerprints, which account for the spatial arrangement of molecular fragments, provides higher prediction accuracy compared to structural keys, which only reflect the presence or absence of certain structural elements. Special attention was paid to predicting the glass transition temperatures of isomeric organic homopolymers, for which accounting for the spatial arrangement of substituents is critical. The results demonstrate the promise of using machine learning methods for predicting polymer glass transition temperatures based on glass transition theories, and indicate the need for further research into similar hybrid models.
Keywords:
MACHINE LEARNING ALGORITHMS, INCREMENTAL METHOD, QSPR MODEL, ORGANIC HOMOPOLYMERS
MACHINE LEARNING ALGORITHMS, INCREMENTAL METHOD, QSPR MODEL, ORGANIC HOMOPOLYMERS
