AI drug discovery depends on better lab data

Drug developers are using AI to move from empirical screening toward predictive design, generating candidates in silico before lab testing. Paul Belcher, director of protein research strategy at Cytiva, says the technology could reduce clinical risk by improving candidate quality, but AI still cannot reliably predict kinetics or developability, leaving wet-lab validation essential.

The shift is increasing demand for high-throughput, information-rich lab systems. Traditional hit identification can screen hundreds of thousands, sometimes millions of compounds with low-fidelity yes-or-no outputs, while AI-generated hits require deeper characterization, purification, and validation.

Better data remains the main constraint. Public datasets often skew toward positive results, missing negative findings that could help models avoid bias, and data integrity is a growing concern: Elisabeth Bik found almost 4% of biomedical papers contained duplicated or manipulated images in 2016.

Autonomous labs could close the loop between AI prediction and physical experiments if instruments become more interoperable and datasets follow FAIR principles. No drug discovered primarily through AI-driven design has yet received full FDA approval, though Belcher expects that to change in the next two to three years.