AI-Designed CRISPR Enzymes Outperform Nature's Best Gene-Editing Tools

For the first time, researchers have used artificial intelligence to design synthetic gene-editing enzymes from scratch — and the AI-created variants work better than the natural versions found in bacteria.

Nobel laureate Jennifer Doudna's lab at UC Berkeley has achieved a milestone in synthetic biology: using AI to design functional CRISPR enzymes that do not exist in nature. The work, published in Science in July 2026, demonstrates that artificial intelligence can now create molecular tools that evolution never produced — and that are more effective than nature's own designs.

The breakthrough: AI-designed SynTnpB nucleases — synthetic variants of the TnpB enzyme, a minimal ancestor of CRISPR-Cas12 — achieved gene-editing activity in human cells that matched or exceeded natural TnpB, with the best variant sharing only 77% of its sequence with the wild-type enzyme.

From natural blueprint to AI design

The team started with TnpB, a compact RNA-guided nuclease that is evolutionarily related to the larger CRISPR-Cas12 enzymes. Using an inverse protein-folding model called ESM-IF1 — originally developed by Meta AI — the researchers fed the desired three-dimensional structure of the nuclease into the model and asked it to generate entirely new amino-acid sequences that would fold into that shape. The result was a library of synthetic variants, which the team named SynTnpBs.

Better than the real thing

When tested in human cells, many SynTnpB variants showed robust gene-editing activity. The most effective variant not only worked but outperformed wild-type TnpB in editing efficiency. Cryo-electron microscopy confirmed that the synthetic enzymes adopted the intended three-dimensional structure. This marks a departure from traditional CRISPR development, which has always relied on discovering and tweaking naturally occurring bacterial enzymes.

What this means for gene therapy

The ability to design functional nucleases on demand opens the door to custom CRISPR tools tailored for specific therapeutic applications. Smaller enzymes are easier to deliver via viral vectors, and AI-designed variants could be optimized for reduced off-target effects, enhanced specificity, or activity in particular cell types. The approach also demonstrates that the CRISPR toolbox is no longer limited to what nature has evolved — biology can now be engineered computationally.

Knowledge takeaway: AI-designed SynTnpB nucleases — synthetic CRISPR enzymes created by inverse protein folding — match or exceed natural gene-editing activity in human cells. The approach, published in Science by the Doudna lab, expands the CRISPR toolbox beyond naturally occurring enzymes and opens the door to custom-designed gene-editing tools for therapy.