Towards Quantitative Structure-Property Relationships for Polymer Biodegradability

Group contribution theories and more recent machine learning methods provide powerful methods to elucidate quantitative structure-property relationships (QSPR) by which the physical properties of polymers may be predicted based on their chemical structure. The increasing focus on sustainability has motivated a desire to incorporate end-of-life considerations into polymer design, necessitating QSPR for polymer biodegradation. However, the extremely limited data available has made this task difficult. Here, we report the synthesis of a large library of polyesters and polycarbonates and the development of a high-throughput biodegradation assay based on the classic clear zone assay. Using these methods, we were able to successfully screen over 640 polyesters for biodegradability, producing a large data set. Analyzing the data for chemical trends reveals key effects of chain length, ring structures, side groups, and heteroatom substitutions within the polyesters. We then applied both logistic regression and random forest classifiers to predict the biodegradation of polyesters, showing a predictive accuracy of up to 82%. Including information on molar mass and the physical state of the polymer did not improve predictions beyond those obtained with chemical structural information alone.