KAIST develops unmanned platform that synthesizes drug candidates around the clock

Researchers have developed a technology that allows robots to perform complex chemical experiments for synthesizing drug candidates in place of human scientists.

KAIST announced Sunday that a research team led by Professor Park Yoon-su of the Department of Chemistry has developed a "24-hour unmanned synthesis platform" combining visible-light photochemistry with continuous-flow processing.

Drug development requires the physical synthesis of molecules — often designed by AI — to verify their efficacy and safety. The process has long posed challenges: even minor changes to a molecular structure demand a new set of synthesis conditions, and scaling up from a small successful batch requires yet another round of adjustments.

To overcome these limitations, the research team integrated a raw-material preparation robot, automated pumps, a photochemical reaction unit and analytical instruments into a single connected system.

Once a researcher sets the initial conditions, the robot automatically mixes the starting materials and pumps the solution through a narrow tube. Visible light is then applied to trigger the chemical reaction, and the resulting solution is fed into analytical instruments that automatically confirm whether synthesis occurred and measure output.

The key innovation is a continuous-flow process that sustains chemical reactions by supplying raw materials without interruption. The narrow tubing allows light to be delivered uniformly throughout the solution, and output can be scaled up simply by extending the duration of operation.

To validate the platform's performance, the research team applied a "single-atom editing reaction" — a technique that selectively replaces one specific atom in a molecule while keeping the rest of its structure intact.

As a result, the team confirmed the feasibility of synthesis across more than 20 furan-series compounds. Even with structurally complex natural products, the platform successfully converted only the targeted portion while leaving the rest of the molecule unchanged.

Particularly notable was the team's success in continuously synthesizing N-methylpyrrole — a core molecular scaffold widely used in pharmaceutical development — demonstrating the platform's potential for scaling up output.

The technology is expected to accelerate the rapid acquisition of diverse drug candidates that differ only in partial molecular structure. It is also anticipated to aid in building "candidate libraries" that allow researchers to compare the efficacy and toxicity of multiple candidates side by side.

The team plans to integrate AI into the platform, developing it into a self-learning system that analyzes accumulated experimental data to independently identify and refine optimal reaction conditions.

"How quickly and reliably we can synthesize a designed molecule will determine the pace of drug development from here on," Park said. "This platform quantifies and reproduces reaction conditions, and will serve as a foundation for the era of automated organic synthesis."

The findings were published Aug. 7 in JACS Au, an international journal of the American Chemical Society.