StoryResearch
Shift Bioscience publishes a calibration method for AI virtual-cell models in Nature Biotechnology
Shift Bioscience published research in Nature Biotechnology, announced on October 5, 2026, on how to score AI 'virtual cell' models that predict cell responses to gene activation or inhibition. The company argues that some reports of weak model performance came from comparison metrics that were miscalibrated and missed real differences between models.
- The work builds on Shift's November 2025 release, which identified metrics that held up across datasets.
- The announcement gives no numerical performance results, and no independent replication is cited.
- Next, Shift plans large-scale in vitro and in silico screens for dual-purpose targets, starting with fibrosis and rejuvenation; SB-101 is its first such target.
- Shift's chief scientific officer says the framework is already used in its target identification program, with a route toward clinical development.