AI predicts DNA repair for safer genome editing

A new AI model can predict DNA repair following CRISPR-based genome editing, suggesting how to achieve more accurate gene insertions.

Scientists from Switzerland and Belgium designed the AI model, named Pythia after an Ancient Greek oracle, to improve the accuracy of gene insertion and reduce unintended effects. Pythia uses machine learning to predict the likelihood of DNA repair based on the sequences surrounding the target cut site and suggests the DNA templates that are most likely to lead to accurate repair.

'Our team developed tiny DNA repair templates, which act like molecular glue and guide the cell to make precise genetic changes,' said Dr Thomas Naert from Ghent University, Belgium, lead author of the study published in Nature Biotechnology.

CRISPR/Cas9 genome editing is used to cut DNA at specific locations, to facilitate adding or removing sections. The broken DNA must then be repaired, and the body has several different mechanisms to achieve this. Of these, homology-directed repair (HDR), is most accurate, but can only happen in cells that are proliferating, meaning it cannot be used in brain or heart tissue.

Nonhomologous end joining (NHEJ) and microhomology-mediated end joining (MMEJ) can be used in these tissues, but are more error-prone and can introduce unintended mutations.

The researchers tested the predictions of Pythia in frog embryos and live mice. Genome editing was successfully performed in mouse brain and heart tissue using MMEJ, resulting in precise edits in non-proliferating, differentiated cells, which represents a major advance over HDR-based methods. DNA repair predicted by Pythia was then validated in 32 different loci in human kidney cells.

The researchers hope that Pythia could help track gene activity and understand disease pathways by aiding the insertion of fluorescent tags into tissues, such as the brain or heart, which were previously inaccessible due to the limitations of non-HDR methods.

However, although the genome editing using Pythia reduced unwanted edits, they were not eliminated. Further refinement and testing will be required before it can be implemented in larger animal models.

'What excites us most is not only the technology itself, but also the possibilities it opens. Pythia brings together large-scale AI prediction with real biological systems. From cultured cells to whole animals, this tight loop between modelling and experimentation points is becoming increasingly useful, for example in precise gene therapies,' said co-author Professor Soeren Lienkamp from the University of Zurich, Switzerland.