Closing the data loop in AI-driven drug discovery

Details

Eroom's Law has driven pharmaceutical R&D costs to double roughly every nine years since the 1950s — today, a single drug costs $1–2.5 billion and takes 10–15 years to reach market, with over 90% of candidates failing in clinical trials. AI is now the industry's primary bet for reversing this trend by compressing discovery timelines and improving candidate quality before expensive clinical phases. The most immediate AI application is in hit identification: instead of physically screening libraries of hundreds of thousands or millions of compounds against a disease target, companies now use AI to design candidates from scratch and predict binding behavior computationally. This removes physical-scale constraints and lets teams eliminate low-quality candidates before committing them to R&D resources, according to Paul Belcher, director of protein research strategy at Cytiva. The practical bottleneck has shifted downstream. AI cannot yet reliably predict compound kinetics or developability, so every AI-generated hit still requires wet-lab validation. Traditional screening workflows optimized for binary yes/no responses at scale are poorly suited to characterizing the more complex and diverse candidates AI generates — creating pressure on lab teams to handle a larger volume of richer, higher-resolution characterization work. Note: This article is sponsored content produced in partnership with Cytiva, a life sciences instrumentation company, which shapes its emphasis on lab characterization infrastructure as the necessary complement to AI candidate generation.