A near-autonomous AI chemist improves a challenging reaction in medicinal chemistry

Advanced AI can be a powerful partner for scientists, accelerating discoveries that benefit humanity. OpenAI has demonstrated this in mathematics and theoretical physics, and by lowering the cost of cell-free protein synthesis. This project extends AI's role into medicinal chemistry, where hypotheses must be validated in labs with real molecules and experimental conditions. OpenAI partnered with Molecule.one, integrating GPT-5.4 with Maria, an AI-driven high-throughput laboratory.

The system was tasked with improving a critical reaction class. GPT-5.4 generated research proposals, designed and executed experiments, analyzed data, and suggested follow-up actions. Humans guided the process by setting prompts, grading proposals, and making minor experimental adjustments. They also validated the final results, ensuring a collaborative approach between AI and human expertise.

The most promising proposal, OAI-M1-03, focused on improving the Chan–Lam coupling, a challenging yet vital reaction for forming carbon-nitrogen bonds. GPT-5.4 autonomously identified primary sulfonamides as a high-value substrate class and surprisingly suggested mild oxidants like TEMPO to boost reaction yields. This finding underscores AI's capacity to uncover non-obvious solutions.

Two cycles of experimentation in Maria Lab confirmed significant improvements. Yields increased for 88% of boronic acids and 83% of sulfonamides tested, with the mean yield rising from 16.6% to 25.2%. Human chemists validated these micro-scale results at bench scale, confirming higher yields for most reactions. This reliability is crucial for drug discovery, where reactions must perform consistently in practical lab workflows.

Improvements in Chan–Lam coupling are particularly impactful because synthesis often bottlenecks drug discovery. Sulfonamides, found in many therapeutic areas like anticancer and antimicrobial drugs, historically show low yields in this reaction. Making the Chan–Lam coupling more reliable provides medicinal chemists a practical way to develop and explore new therapeutic molecules. This project exemplifies AI's potential to revolutionize scientific research by acting as a valuable partner throughout the discovery process.