AI Can Warn Researchers Where CRISPR Might Make a Mistake

Microsoft has built an artificial intelligence tool that predicts the accuracy of CRISPR so that researchers can avoid making incorrect edits in DNA.

Missing the target: CRISPR uses two components: a cutting protein and a guide RNA that directs it to the part of a genome you want to cut. The guide RNA is about 20 letters long. Problem is, multiple sites in a genome can have the same series of letters, so CRISPR could snip the wrong section—known as an “off target” effect.

Why that matters: Off-target effects are one of the biggest safety concerns with CRISPR. Making incorrect cuts in a genome could, say, switch on a cancer-causing gene.

How AI helps: Microsoft’s tool lets researchers plug in a gene they want to modify and get an estimate of how bad potential off-target effects could be. Using the tool, researchers could figure out which genes are most difficult to safely edit and avoid tweaking them, says Nicolo Fusi of Microsoft Research.