MolCluster: An Unsupervised Framework for Multiscale Molecular Representations with Physically Consistent Resolution Control

Coarse-grained (CG) modeling simplifies molecular systems by mapping groups of atoms into representative particles, but traditional approaches depend on fixed rules and struggle to handle diverse chemical structures. Supervised CG methods further suffer from limited labeled datasets and the inability to control mapping resolution, which is essential for multiscale modeling. To overcome these limitations, we propose MolCluster, an unsupervised model that integrates graph neural networks and community detection algorithm to extract CG representations. Additionally, a predefined group pair loss ensures the preservation of target groups, and a bisection strategy enables precise, customizable resolution across different molecular systems. In the case of downstream task, evaluations on the MARTINI2 dataset demonstrate that MolCluster, benefiting from its label-free pretraining strategy, outperforms both traditional clustering and supervised models in CG mapping and bead type prediction. Overall, these results highlight the potential of MolCluster as a base model for customizable and chemically consistent CG mapping, with future applications extending to polymers, proteins, and other complex multiscale systems.