Shift Bioscience today announced the publication of new research in Nature Biotechnology describing an improved calibration framework for deep learning-based genetic perturbation models — the so-called AI virtual cells. The company says it will use the framework to support large-scale in vitro and in silico screens identifying novel inhibition targets for rejuvenation and age-related disease treatment, initially focusing on fibrosis. The announcement was covered by News-Medical on October 5.

Virtual cells on trial

Genetic perturbation models are designed to predict how cells respond at the transcriptomic level to genetic interventions — switching genes on or off — which could make drug-target screening far more scalable than wet-lab work alone. But the field has a credibility problem: previous studies have questioned their reliability, with some models failing to outperform simple baseline approaches.

An abstract neural network dissolving into a cluster of biological cells beside a ruler realigning into a precise measuring scale.
Calibration, not architecture: Shift argues miscalibrated benchmark metrics dulled sensitivity to genuine model performance. Image: AI Frontier Post (generated graphic).

Shift’s study defines a reliable benchmarking system that accounts for both the biological and technical signals in a dataset. Its central finding: underperformance in some past benchmarks may stem from miscalibration of the metrics used to compare models, which reduces sensitivity to genuine model performance. In other words, the yardstick was warped, not necessarily the models. The work builds on foundational research the company reported in November 2025.

From calibration to the clinic

Shift will now apply the framework directly in its target-identification program: large-scale screens for novel, dual-purpose inhibition targets — genes whose inhibition supports both rejuvenation and the treatment of age-related disease. The screens follow the discovery of SB-101, the company’s first dual-purpose target, and will initially focus on fibrosis, described as a key driver of ageing.

Glowing green cells in a petri dish under a microscope beam, with streams of digital data points rising above.
Shift will run large-scale in vitro and in silico screens for dual-purpose targets after the discovery of SB-101. Image: AI Frontier Post (generated graphic).

“Our findings show that by using well-calibrated metrics and the right dataset, virtual cell models can generate biologically meaningful insights. As a result, we can use them with greater confidence to identify promising new targets that are relevant to aging and disease. We are applying this framework directly in our target identification program, focusing on targets whose inhibition can support both rejuvenation and treatment of age-related disease, giving us a clearly defined route towards clinical development.”

Dr Brendan Swain, CSO and founder, Shift Bioscience

The paper is published in Nature Biotechnology. The announcement keeps the quantitative details in the paper itself — for the exact calibration results, the paper is the primary source.