Lecture by Dr. Volker Settels

Polymer Informatics at BASF

Polymers represent an important category of products within BASF. However, integrating polymer data into the data-driven modeling task remains a significant challenge. The primary reason for this challenge lies in the stochastic nature of polymers, which complicates the establishment of a digital, structure-based representation for them. The lack of a structure-based representation for polymers also affects the management of polymer data, including similarity searches and substructure searches. We have identified the absence of such representations as a major technological gap in processing polymer data and have started an activity to address that gap.

An effective polymer representation should encapsulate the chemical structure and the amount of polymer building blocks, along with the bonding probabilities between these units. Additionally, it should adequately convey the stochastic nature of polymers and their microstructures, which may include block, random, branched, and other configurations. In recent years, several representation techniques for polymers have been published in the literature. We have assessed various of those techniques and identified a promising candidate for BASF, which is G-BigSMILES (L. Schneider et al., Digital Discovery 2024, 3, 51).

We are developing G-BigSMILES to address BASF’s specific requirements. One requirement is support for nested stochastic objects, which is necessary for the accurate description of prepolymers and oligomers used as ingredients in the polymerization process. Additionally, the polymer graph description from G-BigSMILES will enable substructure searches for polymer data and facilitate the parameterization of structure-based, data-driven models such as graph neural networks (GNNs).