Imagine being able to precisely edit instructions in our DNA to fix damaged genes and potentially cure diseases. That’s exactly what gene editing with CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats) technology allows scientists to do. CRISPR and CRISPR-associated (Cas) proteins are part of a powerful gene-editing system. These proteins act like molecular scissors that can cut DNA at specific locations, allowing researchers to remove faulty genes, insert new ones, or modify DNA sequences in living cells.

However, CRISPR-Cas9, the most widely used Cas protein variant, can sometimes make mistakes by editing the wrong DNA sequence. These unintended changes, known as off-target cuts, could introduce unwanted mutations that may disrupt healthy genes or lead to unexpected side effects.

Benjamin Kleinstiver, Ph.D., recipient of the NIH Director's High-Risk, High-Reward New Innovator Award, and his team at Massachusetts General Hospital, focused their research on improving the accuracy of CRISPR–Cas9 to develop better, safer, and more precise gene-editing tools. The research team engineered nearly 1,000 variants of the Cas9 gene-editing protein by making small changes to its structure. They tested which DNA sequences each variant recognized and cut, focusing on special sequences called protospacer adjacent motifs (PAMs) which the Cas9 protein needs to work. Using these data, they trained a computer program called “protospacer adjacent motif machine learning algorithm,” or PAMmla, to predict how different Cas9 variants will perform, thereby helping scientists choose the most accurate versions of Cas9. These will ultimately be safer for gene editing in people.

Using this approach, Dr. Kleinstiver and his team designed novel versions of Cas9 proteins that work more precisely and safely in human cells. They also developed AI-powered tools that allow scientists to custom-design Cas9 proteins for specific gene-editing tasks, like correcting disease-causing genes. For example, the researchers designed an approach that selectively targeted a gene with a mutation linked to retinitis pigmentosa, an inherited eye disease that causes blindness. Their approach edited the disease-causing gene without affecting the paired healthy gene in both human cells and mice. This early demonstration shows the possibility of developing highly precise gene-editing tools and is a promising step toward targeted treatments.

Instead of relying on a one-size-fits-all Cas9 protein, which can sometimes make mistakes, this research enables development of custom Cas9 variants that are safer and more effective for specific gene editing tasks. This new method is scalable and efficient; while traditional approaches can only test a few variants, this method leverages AI to evaluate tens of millions of variants.  

Additionally, PAMmla is accessible to the scientific community as a web tool (available at pammla.streamlit.app), allowing researchers across the world to design their own Cas9 variants. This sharing of such an innovative resource has the potential to accelerate development of customizable, safe gene therapies worldwide. Overall, this work marks a significant step toward a future where tailor-made, safe, and efficient gene therapies can be developed for any patient, any disease, anywhere.

Reference: Silverstein, R.A., Kim, N., Kroell, AS. et al. Custom CRISPR–Cas9 PAM variants via scalable engineering and machine learning. Nature 643, 539–550 (2025). https://doi.org/10.1038/s41586-025-09021-y