New study proposes AI-driven framework to refine gene target discovery in rejuvenation research

Cambridge-based Shift Bioscience has proposed a new ranking system for virtual cell models trained on single-cell RNA sequencing (scRNA-seq) data, aiming to improve how researchers evaluate and select tools for gene target discovery.

In a study led by head of machine learning, Lucas Paulo de Lima Camillo, the team identified key flaws in how virtual cell models are typically assessed — finding that many prominent models are outperformed by a simple prediction of the dataset mean. This calls into question the biological relevance of commonly used performance metrics, particularly in the context of perturbation modelling.

Virtual cell models are designed to simulate how gene expression shifts in response to different perturbations, such as gene upregulation or knockdown. These models can dramatically accelerate early-stage discovery by compressing what would be years of lab experiments into virtual screens — but only if they reliably reflect meaningful biological changes.

According to the new paper, experimental variables such as weak perturbations and control bias can distort traditional benchmarks, giving a false sense of model performance. To address this, the team introduced new pre-processing steps, including differentially expressed gene (DEG)-weighted scoring, baseline calibrations, and DEG-aware optimization objectives — collectively helping to prioritize models with stronger predictive power.

“In this research, our team has shown that by focusing on the development of new metrics and baselines, we can more easily identify models that demonstrate strong predictability,” said de Lima Camillo.

“The paper provides foundational data which will enable us to develop more powerful, biologically-useful perturbation models, ultimately accelerating our therapeutic pipeline and helping us to uncover new targets for rejuvenation therapeutics.”

Shift Bioscience is developing therapies aimed at targeting the mechanisms of ageing and age-related disease through AI-guided gene discovery.