Graph-to-Graph Inverse Design of Surfactants under Multiobjective and Structural Constraints

Rapid, sustainable redesign of large functional molecules requires efficient exploration of vast chemical spaces subject to multiple properties and constraints. Graph neural networks have shown strong performance in molecular property prediction, yet their use for inverse design remains relatively underexplored. Meanwhile, chemical language transformers capable of learning long-range relationships have emerged as a competing paradigm. To address this gap, we propose a novel motif-based graph-to-graph framework for the inverse design of surfactants, in which latent representations are optimized to meet target specifications. We demonstrate the framework on a challenging and industrially relevant problem, targeting the critical micelle concentration (CMC), the air-water surface tension at CMC, and the maximum surface excess concentration, while considering constraints on surfactant classes and compatibility between amphiphiles and counterions. We observed that our framework could accurately target multiple properties with greater accuracy than surfactant-property models from the literature. Additionally, compared with our previously proposed SMILES-based transformer, the graph-to-graph framework significantly improves prediction accuracy and inverse design validity and expands the diversity of generated surfactant classes. Overall, these results establish the motif-based graph-to-graph framework as a more accurate, chemically expressive, and industrially applicable technology for the multiobjective inverse design of large functional molecules.