AI Breaks Down Gene-Chemical Barrier as Tencent, Central South University Publish in Cell

A joint research team from Tencent and Central South University in China has developed a new artificial intelligence (AI) model that predicts drug effects using genetic data, signaling a potential shift in the drug development paradigm.

According to biotech media outlet BioOn on the 24th local time, a joint team from the Tencent Institute of Life Sciences and the Department of Computer Science and Engineering at Central South University published a research paper on the “UniPert-G2CP” deep learning framework in the leading academic journal Cell. The paper is titled “UniPert-G2CP Connects Genetic and Chemical Screening From Molecular Representation to Phenotypic Modeling.”

Genetic screening and chemical screening are essential to drug development, but a substantial gap has existed between them. Genetic screening using technologies such as CRISPR gene-editing tools can examine thousands of genes at once, whereas chemical screening using compound libraries can examine only a small number of molecules because of experimental limitations.

The “UniPert-G2CP” developed by the research team is a two-stage deep learning framework that bridges this gap. The first stage, “UniPert,” serves as a “translator” that maps genes and chemical molecules into the same vector space. This enables direct comparison and connection of data in different forms.

The second stage, “G2CP,” uses a transfer learning strategy. It adds limited chemical screening data to a model pretrained on extensive genetic perturbation screening data. This approach is based on the assumption that a particular drug can act on shared signaling pathways and mimic genetic effects.

The research team validated the model using 4994 genes and 7860 small-molecule compounds across 4 cancer cell lines. The team said the model revealed cellular heterogeneity underlying drug responses and provided insights into mechanisms of drug action and resistance.

However, the research team also specified that the model’s predictions must be validated through actual experiments and that prediction reliability may be lower for compounds with completely new mechanisms of action.

Nevertheless, “UniPert-G2CP” is being viewed as opening a path to using extensive genetic data for virtual drug screening. It is expected to become a key technology for reducing the enormous time and cost of drug development and accelerating the era of precision medicine driven by mechanisms rather than data alone.