Cambridge, England — Shift Bioscience has published new research in Nature Biotechnology describing an improved framework for evaluating deep learning-based genetic perturbation models, a type of AI virtual cell used to predict how cells respond to genetic changes.
The biotechnology company said the framework will support large-scale laboratory and computational screening to identify potential therapeutic targets for rejuvenation and age-related diseases, initially focusing on fibrosis.
Genetic perturbation models are designed to predict transcriptomic responses when genes are activated or inhibited. Such models could enable large-scale virtual target screening, but previous studies have raised questions about their reliability.
Shift said its new framework accounts for biological and technical signals within datasets to provide a more accurate assessment of model performance.
The study found that some previously reported underperformance may have resulted from poorly calibrated benchmarking metrics that were less sensitive to genuine model improvements.
Shift plans to use the framework in large-scale in vitro and in silico screens to identify inhibition targets that could have applications in both cellular rejuvenation and treatment of age-related disease.
The work follows the discovery of SB-101, the company’s first dual-purpose target. Future screening will initially focus on fibrosis, which is associated with aging and multiple chronic diseases.
“Our findings show that well-calibrated metrics and the right datasets can help virtual cell models generate biologically meaningful insights,” said Brendan Swain, Ph.D., Chief Scientific Officer and Founder of Shift Bioscience.
He said the company is applying the framework directly to its target identification program to identify potential therapies for both rejuvenation and age-related disease.
