AlphaFold Repurposed to Trim CRISPR's Off-Target Errors

Researchers in China modified AlphaFold to spot why gene editors cut the wrong DNA, then engineered safer Cas9 variants.

A research team based at multiple institutions in China modified the AI protein-folding software AlphaFold to hunt down the structural quirks behind off-target gene edits. The work, published in Nature, focused on the Cas9 protein, which pairs a guide RNA against roughly 18 bases of genome. In theory an 18-base match should appear once in about 70 billion bases, far more than the 3 billion in the human genome, but Cas9 tolerates a few mispaired bases, and those mismatches drive unwanted cuts.

To map the problem, the team built a library of off-target sites using a modified CRISPR system that converts adenine to inosine, repeating the process across 10 different guide RNAs. They found that about two-thirds of off-target sites made Cas9 adopt a slightly different shape, and over 95 percent changed which amino acids touched the RNA. Their analysis method, named ContactSeek, guided 23 amino acid swaps at 10 key positions to find safer Cas9 variants.

Safer, more predictable editing is a foundation for making gene therapy cheaper and more widely deployable. Fewer off-target cuts means less risk to validate, which lowers the cost and time of bringing a genetic medicine to patients.

The caveats matter. AlphaFold could not correctly model the full multi-protein complex and only worked when fed a simplified input of DNA, RNA, and Cas9 alone. The results are computational analysis checked against some experimental structures, not clinical validation. Whether the engineered variants hold up in living systems is the next question.

Source: Ars Technica