Polymer properties and activity coefficients for mechanistically informed AI for polymer networks

Polymer networks are ubiquitous, and are utilized for a wide range of applications from commodity rubbers to specialized drug delivery applications. While structure-property relationships in model polymer network systems have been extensively studied recently, a major challenge in the discovery of new optimized networks for novel applications is finding the molecules that produce new emergent properties within a vast design space. Experimental measurements are costly, and innovation could be accelerated by a three-way partnership between data science, theory, and experiment. Predicting macroscopic properties of networks for novel applications requires the knowledge of properties of the underlying polymer chains. Key thermodynamic properties of polymers, such as activity coefficients, are extremely important to predict non-ideal behavior in mixtures. In this work, a database of three sets of crucial polymer properties relevant for polymer network mechanics and solution phase behavior was curated. A combination of quantum calculations, atomistic simulations and thermodynamic solvation models was utilized to compute statistical segment lengths, activity coefficients in various solvents, cleavage pattern and relevant energetics. This database serves as a platform for utilization in machine learning algorithms to drive the prediction of network properties for novel applications.