AI enables precise gene editing

A research team led by the University of Zurich (UZH) has developed a powerful new method to precisely edit DNA by combining cutting-edge genetic engineering with artificial intelligence (AI).

Precise and targeted DNA editing by small point mutations as well as the integration of whole genes via CRISPR/Cas technology has great potential for applications in biotechnology and gene therapy. However, it is important that the so-called gene scissors do not cause any unintended genetic changes, but instead maintain genomic integrity in order to avoid unintended side effects.

Scientists from UZH, Ghent University and ETH Zurich have now created a method which utilities AI to greatly improve the precision of genome editing. Dubbed ‘Pythia’ and described in the journal Nature Biotechnology, the tool predicts how cells repair their DNA after it is cut by gene editing tools such as CRISPR/Cas9.

“Our team developed tiny DNA repair templates, which act like molecular glue and guide the cell to make precise genetic changes,” said lead author Thomas Naert, who pioneered the technology at UZH and is currently a postdoc at Ghent University. These AI-designed templates were first tested in human cell cultures, where they enabled highly accurate gene edits and integrations; the approach was also validated in other organisms, including Xenopus, a small tropical frog used in biomedical research, and in living mice, where the researchers successfully edited DNA in brain cells.

Pythia is named after the high priestess of the oracle at the Temple of Apollo at Delphi in antiquity, who was consulted to predict the future. In a similar way, this new tool allows scientists to forecast the outcomes of gene editing with remarkable precision.

“Just as meteorologists use AI to predict the weather, we are using it to forecast how cells will respond to genetic interventions,” said senior author Soeren Lienkamp, a professor at the Institute of Anatomy of UZH and ETH Zurich. “That kind of predictive power is essential if we want gene editing to be safe, reliable and clinically useful.”

This is possible because DNA repair follows patterns — it is not random — so when CRISPR cuts DNA, scientists typically rely on the cell’s natural repair mechanisms to fix the break. But these repairs can result in unwanted outcomes, such as destruction of the surrounding genes.

“What we modelled at massive scale is that this DNA repair process obeys consistent rules that AI can learn and predict,” Naert said. With this insight, the researchers simulated millions of possible editing outcomes using machine learning, asking a simple question: what is the most efficient way to make a specific small change to the genome, given how the cell is likely to repair itself?

In addition to changing individual letters of the genetic code or integrate an exogenously delivered gene, the method can also be used to fluorescently label specific proteins. “That is incredibly powerful, because it allows us to directly observe what individual proteins are doing in healthy and diseased tissue,” Naert said.

Another advantage of the new method is that it works well in all cells — even in organs with no cell division, such as the brain. Indeed, it opens the door to a range of possibilities — from more accurate modelling of human diseases to the development of next-generation gene therapies.

Image caption: Fluorescently tagged neural molecule imaged in a living tadpole, with colours representing imaging depth. The brain and spinal nerves appear near the top in turquoise to purple; the path of peripheral nerves is visible throughout the tadpole. Image credit: Taiyo Yamamoto, UZH.