MolLingo: Multi-Agent System for AI-Driven Molecular Design
MolLingo functions as a multi-agent system that streamlines molecular design by mimicking the thought processes of chemists. It operates through a shared memory module that links a Literature Agent, a Chemist Agent, and an Orchestrator, each utilizing specialized tools for their domains. The system features BRICS-based Fragment Enumeration (BFE), a method for molecular fragmentation that is synthesis-aware, breaking down molecules into block-based SMILES alongside standard chemical names. This innovation connects molecular structures with LLM semantic spaces, facilitating block-level reasoning and modifications. By doing so, it overcomes the shortcomings of current LLM-based approaches, which often lack coordinated multi-agent efforts and iterative, evidence-based reasoning.
Key facts
- MolLingo is a multi-agent system for molecular design.
- It emulates the reasoning process of a chemist.
- The system includes a Literature Agent, a Chemist Agent, and an Orchestrator.
- Agents share a memory module and have domain-specific tools.
- BFE is a synthesis-aware molecular fragmentation method.
- BFE decomposes molecules into block-based SMILES with common chemical names.
- The representation bridges molecular structure and LLM semantic space.
- Existing LLM approaches lack multi-agent coordination and iterative reasoning.
Entities
—