AI Spots Gene-Editing Mistake Makers, Paving Way for Safer Therapies
Scientists have used artificial intelligence to identify why CRISPR sometimes cuts the wrong DNA, leading to a new method that reduces dangerous errors in gene editing.
Researchers are using Google's AI tool, AlphaFold, to pinpoint the specific parts of gene-editing proteins that cause off-target cuts [204515]. By analyzing how CRISPR complexes interact with DNA, the AI model reveals why the system occasionally slices unintended sections of the genome—a major barrier to safe medical use [202261]. This allows scientists to predict and avoid these mistakes, making gene therapy safer for patients [204515].
The breakthrough builds on a new technique called AlphaFold3-based contact modelling, which predicts how editing proteins bind to DNA [202243]. The AI maps the exact contact points between the protein and the DNA strand, enabling precise changes to a single “letter” in the genetic code without disrupting nearby genes [202243]. This approach reduces unintended mutations, a common problem with traditional CRISPR tools that can cut in the wrong place [202243].
Separately, researchers have developed “AI-CRISPR,” a system that uses machine learning to design guide RNAs—molecules that direct CRISPR to its target [201041]. The AI predicts which guide RNAs will work best, cutting errors and off-target effects [201041]. In lab tests, the upgraded tool edited genes with higher accuracy than standard CRISPR, with the lead author stating, “AI helps us avoid trial and error, making gene therapy safer and faster” [201041]. The team plans to test AI-CRISPR in animal models, with human trials possible within three years [201041].
These advances bring CRISPR closer to reliable, error-free editing for treating genetic diseases like sickle cell anemia and cystic fibrosis [201041][202261].