Shift Bioscience publishes AI virtual cell calibration framework in Nature Biotechnology

Shift Bioscience has published a peer-reviewed calibration framework to address miscalibration in deep learning genetic perturbation models, aiming to increase the reliability of AI-driven screens for cellular rejuvenation and fibrosis targets.

Shift Bioscience announced the publication of a study in Nature Biotechnology describing an improved calibration framework for deep learning-based genetic perturbation models, also known as AI virtual cells.

The peer-reviewed research demonstrates that past model underperformance in benchmarks often stemmed from miscalibrated metrics rather than fundamental model failure, leading to a revised benchmarking system that considers both biological and technical signals.

The company reported that it will use the calibrated framework to launch large-scale in vitro and in silico screens to identify dual-purpose inhibition targets that address both cellular rejuvenation and the treatment of fibrosis.

The initial application of these screens will focus on expanding the company's platform beyond its first dual-purpose target, SB-101.

AI virtual cells are designed to simulate transcriptomic responses to genetic intervention, offering a scalable alternative to physical screens, but their real-world utility has been constrained by concerns over baseline predictive accuracy.

Establishing a validated benchmarking standard could provide a more reliable path for evaluating computational drug-discovery models, aiding researchers in identifying genuine biological signals over background noise.

By focusing on dual-purpose targets, the platform aims to systematically couple cellular rejuvenation mechanisms with concrete clinical pathways, specifically targeting liver and organ fibrosis.

Within the AI-driven drug discovery sector, the publication of standardized, peer-reviewed evaluation frameworks is key to building clinical and institutional confidence in computational target discovery.

If Shift Bioscience successfully translates these calibrated virtual predictions into validated in vitro candidates, it could support the commercial viability of using narrow AI platforms to target complex, multi-system processes like epigenetic aging.

The launch and initial screening results from Shift Bioscience's large-scale in silico and in vitro targeting campaigns.

Preclinical progress of the company's lead program in liver fibrosis, which aims to validate the platform's ability to discover candidates that reverse epigenetic age while resolving disease pathology.