MolSets: Molecular Graph Deep Sets Learning for Mixture Property Modeling (2312.16473v1)

Abstract: Recent advances in ML have expedited materials discovery and design. One significant challenge faced in ML for materials is the expansive combinatorial space of potential materials formed by diverse constituents and their flexible configurations. This complexity is particularly evident in molecular mixtures, a frequently explored space for materials such as battery electrolytes. Owing to the complex structures of molecules and the sequence-independent nature of mixtures, conventional ML methods have difficulties in modeling such systems. Here we present MolSets, a specialized ML model for molecular mixtures. Representing individual molecules as graphs and their mixture as a set, MolSets leverages a 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 conductivity of lithium battery electrolytes and highlight its benefits in virtual screening of the combinatorial chemical space.

Introduction

The development of new materials, such as those used in battery electrolytes, often hinges on understanding complex structure-property relationships. Machine Learning (ML) has emerged as a powerful tool to unravel these relationships, offering efficient means to navigate the immense chemical space of potential materials. However, challenges arise when dealing with molecular mixtures due to their inherently vast combinatorial nature and sequence-independent configurations.

MolSets Model

To address the unique challenge of accurately modeling the properties of molecular mixtures, a new ML model, MolSets, has been introduced. This model approaches molecular mixtures from a novel perspective: it treats them as sets of molecular graphs. By adopting this approach, MolSets can embrace the complexity at the molecular level, allowing for detailed consideration of individual molecular chemistry and geometry, while also considering the global permutations of constituents within mixtures.

MolSets combines three key components: a graph neural network (GNN) for processing the molecular information, an attention mechanism to weight the significance of constituent interactions, and the Deep Sets architecture to ensure permutation invariance. This combination enables MolSets to extract and aggregate complex molecular information to predict the properties of a molecular mixture accurately.

Model Demonstration

MolSets’ effectiveness was exemplified by predicting the conductivity of lithium battery electrolytes. The model was tailored to accommodate molecular mixtures consisting of various solvents and a single common salt, capturing the interaction of these compounds at room temperature. MolSets outperformed other ML models by adhering to the set-based representation that respects the sequence-independent nature of mixtures. This adherence ensures that different permutations of the same constituents are viewed as similar, leading to more accurate predictions.

Practical Application and Outlook

MolSets presents a promising avenue for the virtual screening of molecular mixtures, expediting the discovery and design of new materials within the expansive combinatorial space. By applying MolSets to existing datasets, the model can virtually predict the properties of thousands of potential new electrolytes, facilitating the identification of promising candidates that may offer improved performance.

Moving forward, expanding databases to include mixture properties and implementing models like MolSets on collaborative platforms could drive the development of better materials. These efforts would parallel advancements in the prediction of protein structures and crystal properties, showcasing the transformative impact of ML in materials research.