CGMas: Multi-Agent AI Framework Automates Polymer Coarse-Grained Modeling
A novel multi-agent system called CGMas has been introduced in a preprint available on arXiv (arXiv:2608.06694), which streamlines the creation of coarse-grained (CG) molecular dynamics models for polymers. Utilizing a large-language-model (LLM) reasoning agent, it deduces the all-atom (AA) topology from a natural-language description of the polymer and its desired resolution. The framework employs a layered self-correction process to address physical inaccuracies often found in unsaturated, heteroatom-rich, and polar polymers. Additional agents manage tasks such as equilibration, CG representation mapping, potential derivation via Boltzmann inversion, and benchmarking against atomistic references. CGMas successfully navigated all 27 test scenarios, showcasing its potential to automate the traditionally labor-intensive CG modeling process. This advancement could greatly enhance polymer simulation research, facilitating quicker investigations of polymer characteristics beyond all-atom approaches. The preprint was released as new on arXiv with the identifier 2608.06694v1.
Key facts
- CGMas is a multi-agent framework for automated coarse-grained molecular dynamics of polymers.
- It uses a large-language-model (LLM) reasoning agent to infer all-atom topology from polymer name.
- Layered self-correction resolves physical errors in unsaturated, heteroatom-containing, and polar polymers.
- Downstream agents automate equilibration, mapping, potential derivation, and validation.
- Potentials are derived via Boltzmann inversion.
- CGMas completed all 27 test cases.
- The framework automates bottom-up CG modeling, which is typically laborious.
- The preprint is available on arXiv with identifier 2608.06694v1.
Entities
Institutions
- arXiv