AI starts designing the molecular scissors behind gene therapy

Researchers used AI to design and refine CRISPR editors with higher fidelity, including nucleases built largely from scratch.

Two research efforts used artificial intelligence to build better CRISPR gene-editing tools. CRISPR relies on nuclease proteins, steered by guide RNA, to cut and alter DNA, and a common problem is bystander editing, where the tool changes neighboring letters it should leave alone. AI is now being aimed at that precision problem.

One Chinese team used Google DeepMind's AlphaFold 3 to estimate how parts of CRISPR proteins interact with DNA, then built a tool called ContactSeek to flag protein regions tied to mistaken targeting. Using it, they improved an A-to-G base editor with just two changes, and it outperformed several existing high-fidelity editors. Separately, a group including Jennifer Doudna used AI to design entirely new compact nucleases based on Cas12; the synthetic versions differed from natural ones by about 30 percent, worked in bacterial, plant, and human cells, and a few beat their natural counterparts on efficiency.

Gene therapy has been extraordinarily expensive to develop. Tools that are more accurate and can be designed computationally shorten the discovery cycle and compound the long decline in biotech costs, which is what eventually brings treatments within reach of more patients.

The honest limits are real: these editors still need to be tested inside living bodies, researchers must confirm the synthetic proteins do not provoke an immune attack, and ContactSeek is only as good as its training data. Neither study directly tackled bystander editing, and the whole field is early-stage.

Source: Singularity Hub