Updated
Updated · MIT Technology Review · Jul 27
Cytiva Flags 90% Drug Failure Rates as AI Discovery Hits Data, Integration Walls
Updated
Updated · MIT Technology Review · Jul 27

Cytiva Flags 90% Drug Failure Rates as AI Discovery Hits Data, Integration Walls

1 articles · Updated · MIT Technology Review · Jul 27

Summary

  • Cytiva said AI drug discovery is being held back by weak data quality and poorly integrated lab systems, even as the industry looks to cut 10-15 year development timelines and $1 billion-$2.5 billion costs.
  • AI is helping companies move from physical screening to predictive design, generating more candidate hits earlier, but those compounds still need lab validation because models cannot yet reliably predict kinetics or developability.
  • Public datasets are becoming a bottleneck because they lack structure, diversity and negative results, while publication bias leaves failed experiments buried and limits models' ability to learn what not to pursue.
  • Data integrity is a second risk: Cytiva cited research finding nearly 4% of biomedical papers contained duplicated or manipulated images, and pointed to image-checking tools using secure hashes to detect tampering.
  • Cytiva sees autonomous 'dark labs' as the longer-term goal, but says most labs still run standalone instruments; no primarily AI-designed drug has full FDA approval yet, though it expects that within 2-3 years.

Insights

Could AI-generated drug candidates actually slow down the discovery process by overwhelming traditional wet labs with thousands of untested molecules?
If AI requires learning from failures, how can the secretive pharma industry be convinced to share its costly negative data?
As fully autonomous dark labs become reality, will human intuition in biological research be entirely replaced by continuous algorithmic testing?