G2RINS: A Generative String-and-Graph Polymer Representation to Assist Computational Materials Discovery

<jats:p>There is considerable interest in developing generative models for the design of new polymeric materials. A major challenge towards the development of such models is designing machine-readable representations that capture the hierarchical and stochastic nature of polymers. In this work, we introduce G2RINS, a generative line and graph notation that builds on G-BigSMILES to account for the multistep hierarchical microstructure of real polymers by properly describing prepolymers and branching. This new string representation considerably improves the descriptive capabilities of the overall framework. Additionally, our dedicated Python implementation automatically converts G2RINS strings into generative graphs to create meaningful polymer chain ensembles. Together, the G2RINS string, graph, and generated ensembles serve as versatile tools for database labeling, high-throughput molecular dynamics (MD) simulations, machine learning (ML) modeling, and fingerprinting. Ultimately, G2RINS provides a multilevel, machine-readable representation with the potential to dramati-</jats:p>