Closed-Loop Reinforcement Learning in De Novo Molecular Design: Bridging Generative AI with Automated Chemical Synthesis
Abstract: Closed-loop reinforcement learning offers a promising approach to integrating generative molecular design with automated chemical synthesis and experimental evaluation. However, translating autonomous chemical optimization into pharmaceutical discovery remains challenging because biological measurements are often delayed, variable, and affected by assay limitations. This critical review and perspective examines current autonomous discovery platforms and evaluates the challenges associated with incorporating physical experimental feedback into molecular optimization. It proposes an uncertainty-aware, multi-objective reward architecture and a four-gate qualification framework covering operational synthesizability, analytical identity and purity, biological assay validity, and governed policy updating. Particular attention is given to distinguishing statistically censored measurements from pharmacologically confounded or invalid observations. The article also introduces supervised autonomy as a practical governance approach in which automated systems perform routine experimental operations while scientific experts retain responsibility for critical decisions. The proposed framework provides a foundation for improving the reliability, interpretability, and practical implementation of experimentally informed molecular discovery. Prospective validation remains necessary to establish its effectiveness.
Keywords: Generative artificial intelligence; reinforcement learning; de novo molecular design; automated chemical synthesis; pharmaceutical developability
How to Cite?: Michele Covington, "Closed-Loop Reinforcement Learning in De Novo Molecular Design: Bridging Generative AI with Automated Chemical Synthesis", Volume 15 Issue 10, October 2026, International Journal of Science and Research (IJSR), Pages: 1389-1396, https://www.ijsr.net/getabstract.php?paperid=SR261014212517, DOI: https://dx.doi.org/10.21275/SR261014212517