Shift Bioscience publication increases confidence in AI virtual cells for novel target discovery
Nature Biotechnology publication details improved calibration framework for deep learning-based genetic perturbation models
6 Oct 2026
Shift Bioscience (Shift) has announced the publication of new research describing an improved calibration framework for deep learning-based genetic perturbation models1.
Genetic perturbation models, a type of AI virtual cell, are designed to predict how cells respond at a transcriptomic level to genetic interventions, including activation and inhibition of genes. These models can support scalable in silico target screening, but previous studies have questioned their reliability, with some models failing to outperform simple baseline approaches.
The study from Shift Bioscience builds on foundational research reported by the Company in November 20252. It defines a reliable benchmarking system for models that considers the biological and technical signals in a dataset, providing more meaningful insight into model performance.
The framework demonstrated that model underperformance in some past benchmarks could stem from miscalibration of the metrics used to compare them, causing reduced sensitivity to genuine model performance.
Shift will now use the findings of the study to launch large-scale in vitro and in silico screens for novel, dual-purpose inhibition targets. Following the discovery of SB-101, Shift’s first dual-purpose target, the screens will focus initially on uncovering targets for both rejuvenation and the treatment of fibrosis, a key driver of aging and age-related disease.
Dr Brendan Swain, CSO and Founder, Shift Bioscience, said, “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.”
References
1. Miller, HE, Mejia, GM, Leblanc, FJA et al. Deep learning perturbation models can outperform baselines on calibrated metrics. Nat Biotechnol (2026). https://doi.org/10.1038/s41587-026-03307-w
2. Miller HE, Mejia GM, Leblanc FJA, et al. Deep Learning-Based Genetic Perturbation Models Do Outperform Uninformative Baselines on Well-Calibrated Metrics. bioRxiv 2025.10.21; doi: https://doi.org/10.1101/2025.10.20.683304
Want the latest science news straight to your inbox? Become a SelectScience member for free today>>
Tags
Frequently asked questions
Show frequently asked questions
What calibration framework did Shift Bioscience develop for deep learning genetic perturbation models?
Shift Bioscience developed a benchmarking framework that accounts for biological and technical signals within datasets. The calibrated metrics provide more meaningful assessments of AI virtual cell performance and showed that some previous underperformance may have resulted from miscalibrated comparison metrics.
How can AI virtual cell models support genetic target discovery?
Genetic perturbation models predict transcriptomic cellular responses to interventions such as gene activation and inhibition. With well-calibrated metrics and suitable datasets, these deep learning models can enable scalable in silico screening and generate biologically meaningful insights for identifying targets related to aging and disease.
What targets will Shift Bioscience investigate following the discovery of SB-101?
Shift Bioscience will conduct large-scale in vitro and in silico screens for novel dual-purpose inhibition targets. Following SB-101, its first dual-purpose target, the company will initially seek targets that support both rejuvenation and fibrosis treatment, addressing a key driver of aging and age-related disease.