Reproduce “MARCO: Multi-Round Agentic Reinforcement for Conditional Molecular Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shicheng Fang, Yuxin Wang, Zhuo Yang, Xiaohu Xu, Jiahao Lu, Chuanyuan Tan, Tong Zhu, Yining Zheng and 1 more
MARCO trains molecular editors through bounded cycles of molecule proposals, verifier feedback, and revision. This lets the policy use structured property and similarity signals rather than treating optimization as a one-answer task. On benchmark evaluators, its first response achieves the highest reported SR×Sim across 18 primary task–split–backbone cells; the score does not establish laboratory or clinical effectiveness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.