AI-powered "virtual cells" that predict how genes behave when activated or inhibited have enormous potential for drug discovery — but their reliability has been a sticking point. Shift Bioscience is tackling that head-on. The biotech published a new study in Nature Biotechnology introducing an improved calibration framework for deep learning-based genetic perturbation models, showing that past underperformance was largely due to poorly calibrated benchmarks rather than the models themselves.

With this more trustworthy framework in hand, Shift is now launching large-scale in vitro and in silico screens to identify novel drug targets — specifically ones that could serve a dual purpose: supporting cellular rejuvenation and treating fibrosis, a major driver of age-related disease. This builds on their earlier discovery of SB-101, their first dual-purpose inhibition target.

Key Takeaways:

• Previous benchmarks may have unfairly dismissed virtual cell models due to miscalibrated metrics
• The new framework enables more meaningful, biologically relevant model comparisons
• Initial screens will focus on fibrosis and aging — two closely linked, underserved areas

Why it matters: Better-calibrated AI models mean faster, more confident identification of drug targets — potentially accelerating therapies for aging-related diseases that affect millions globally.