Researchers have started using the latest version of Google DeepMind's AlphaFold AI to redesign the proteins that do the cutting in gene editing. The new framework, called ContactSeek, is built around AlphaFold3 and targets a major hurdle in the field: accidental off-target edits that can cause dangerous side effects.
What ContactSeek Does
Gene-editing tools like CRISPR rely on a protein that snips DNA at a precise spot. But that protein sometimes cuts in the wrong place. ContactSeek uses AlphaFold3's ability to predict protein structures and interactions to engineer versions of the cutting protein that recognize only the intended genetic sequence. The goal is a protein that simply won't bind to similar-but-wrong DNA.
The team behind ContactSeek trained the AI system on a large library of protein-DNA interactions. AlphaFold3 then suggested mutations that would make the protein more selective. Early tests suggest the redesigned proteins are far less likely to cause off-target cuts, the researchers reported.
Why Safety Matters for Gene Therapy
Unwanted cuts are more than a scientific annoyance. They can activate cancer genes or disrupt healthy cells. That risk has slowed approval of several promising gene therapies. A safer cutting protein could remove a key regulatory hurdle and open the door to treating more diseases, especially in the liver and blood where many current trials are focused.
“The safety problem has been the single biggest obstacle in moving gene editing from the lab to the clinic,” the lead researcher noted in a statement. “If we can design proteins that only cut where we want them to, we remove that obstacle.”
The Investor Angle
For biotech investors, this matters because safer gene editing could revive interest in platforms that had been sidelined by safety concerns. Companies developing in vivo gene-editing therapies have seen their valuations buffeted by every piece of safety data. A reliable method to design safer proteins could reduce that uncertainty.
The ContactSeek framework itself isn't a product yet. But it points to a future where AI predicts not just protein structure but also the exact mutations needed to improve function. Investors are watching whether the team behind ContactSeek will license the technology or start a spinout company.
“The ability to rationally design a safer editor is the kind of breakthrough that changes the math for an entire pipeline,” said a biotech analyst who follows the sector.
What Comes Next
The researchers plan to test the redesigned proteins in animal models over the next year. If those experiments hold up, human trials could follow in two to three years. The biggest question: will the redesigned proteins work as well in living organisms as they do in computational models?
