Constrained Flow Optimization via Sequential Fine Tuning for Molecular Design

Researchers have introduced two new methods, Constrained Flow Optimization (CFO) and CLIO, to improve molecular design through generative foundation models.

CFO is an algorithm that balances reward maximization and constraint satisfaction in generative optimization, providing convergence guarantees for constrained generative optimization and generation^[1]. It achieves consistent increases in reward while ensuring high constraint satisfaction. CLIO, on the other hand, is an agent that couples a continuously-updated belief-state graph with a recursive plan-then-act loop, allowing it to adapt its strategy when its tools or assumptions fail^[2]. In a closed-loop human-AI campaign to design an aqueous organic redox flow battery (AORFB) negolyte, CLIO converged on a top phosphonate candidate with a 130~mV improvement in redox potential over the literature baseline^[2]. The resulting compound showed substantially improved electrochemical reversibility and maintained a 90~mV improvement in redox potential. While CFO provides a rigorous framework for Constrained Generative Optimization, CLIO demonstrates the potential for AI agents to drive scientific discovery in molecular design.