Data bottlenecks slow AI-driven drug discovery
As AI speeds drug candidate design, gaps in data quality, access to negative results, and lab integration determine whether AI can meaningfully shorten development timelines and reduce costs.
At a glance
- Cost to bring a new drug to market is cited as $1 billion to $2.5 billion, with development times of 10–15 years and failure rates above 90%.
- The field cites Eroom’s Law, noting costs have roughly doubled every nine years since the 1950s.
- AI is shifting workflows from empirical screening to predictive design, enabling AI-designed candidates and predicted interactions before physical RD work begins.
- AI cannot yet reliably predict kinetics or developability of new compounds, so all AI-generated candidates require lab validation before further development.
- Data quality and accessibility problems are acute: publicly available datasets are biased toward positive results, and negative data are scarce; some researchers call for a “journal of negative data.”
- Fabrication and other practicalities enabled by AI are creating new concerns about data integrity and reproducibility.
The story
MIT Technology Review examines how AI is accelerating drug discovery but reveals that the speed comes with tangible bottlenecks in the lab and data systems. As AI helps identify new therapeutics targets faster, the main costs now hinge on the clinical phase, and getting reliable data remains essential to translating AI insights into successful drugs.