Researchers have used AlphaFold to help redesign gene-editing proteins in an effort to reduce unwanted DNA changes. Ars Technica reports that a team described the work in Nature, focusing on how AI-predicted structures can reveal parts of CRISPR machinery that allow off-target edits.
Gene editing depends on guide RNA, a Cas protein such as Cas9, and a component that changes DNA. Even when a guide RNA is designed for one target, similar sequences elsewhere in the genome can sometimes be edited by mistake. Those rare errors matter because therapies may need to edit many cells.
The researchers built a library of off-target editing sites, then used AlphaFold to compare how Cas9 interacted with matched and mismatched DNA-RNA structures. The first model setup failed when an added editing enzyme landed in the wrong place, so the team simplified the system to DNA, RNA, and Cas9. That produced structures consistent with experiments and helped identify regions involved in unwanted tolerance. The work does not make gene editing risk-free, but it shows how protein-structure AI can guide narrower, testable safety improvements.