MolSets: Molecular graph deep sets model for mixture property modeling

Recent advances in machine learning (ML) have expedited materials discovery and design. One significant challenge faced by ML for materials is the expansive combinatorial space of potential materials formed by diverse constituents and their possible configurations. This complexity is particularly evident in molecular mixtures, a space frequently explored in search of electrolytes, fuels, and coolants, among other applications. Due to the complex structures of molecules and the permutation-invariant nature of mixtures, conventional ML methods have difficulties in modeling such systems. In this work, we propose MolSets, a specialized ML model for molecular mixtures. Representing individual molecules as graphs and their mixture as a set, MolSets leverages graph neural network and the deep sets architecture to extract information at the molecule level and aggregate it at the mixture level, thus addressing local complexity while retaining global flexibility. We demonstrate the efficacy of MolSets in predicting the transport properties of lithium battery electrolytes, and highlight its broader applicability across various mixture property modeling tasks and other problems with combinatorial complexity.