Enhancing α- and β-glucan esters’ material selection through machine learning: An empirical study

Abstract

Polysaccharide esters, with their potential as biomass plastics, represent sustainable alternatives to oil-based plastics. This study contributes to the optimization of material design by demonstrating that Materials Informatics (MI), combined with machine learning, can be effectively utilized to predict and enhance the properties of polysaccharide esters. The research methodology involved generating Simplified Molecular Input Line Entry System (SMILES) representations for polysaccharide esters, creating a novel dataset from scratch. By employing fourteen distinct machine learning models, the research successfully constructed a Quantitative Structure-Property Relationship (QSPR) model that accurately predicts $T_g$ and Elongation at Break of the given esters. Additionally, the study applied multiobjective optimization to these models, optimizing for both $T_g$ and Elongation at Break. This approach enables the efficient achievement of new material properties by significantly reducing the number of required experiments. The practical application of these models was further validated through laboratory experiments involving the synthesis and testing of proposed polysaccharide ester structures.