AI-Powered Protein Engineering for Precise and Scalable Gene Therapy

Genome editing has advanced rapidly, offering real hope for treating genetic disorders—but there’s still room for refinement. A new study from researchers at Mass General Brigham, published in Nature, introduces a major step forward: the integration of scalable protein engineering and machine learning to enhance the precision, safety, and versatility of gene and cell therapies.

At the heart of the study is a machine learning algorithm called PAMmla, designed to predict the properties of an astounding 64 million CRISPR-Cas9 enzymes. This innovation opens the door to customizing enzymes for therapeutic applications while minimizing off-target effects.

“Our study is a first step in dramatically expanding our repertoire of effective and safe CRISPR-Cas9 enzymes,”
 

Ben Kleinstiver, PhD, Corresponding Author and Founding Member, Mass General Brigham

Kleinstiver, also a Kayden-Lambert MGH Research Scholar and Associate Investigator at Massachusetts General Hospital, notes that the study demonstrated the use of PAMmla-predicted enzymes to precisely edit disease-causing sequences in both primary human cells and a mouse model.

Tackling CRISPR’s Limitations with Machine Learning

CRISPR-Cas9 technology allows scientists to cut and modify DNA at specific sites in the genome, but traditional Cas9 enzymes often carry the risk of off-target effects, unintended edits to non-target DNA regions. This not only impacts the therapy's safety but also limits the range of diseases that can be effectively treated.

The PAMmla platform addresses this by predicting the protospacer adjacent motifs (PAMs), short DNA sequences that Cas9 enzymes must recognize to make a cut. Accurate PAM recognition is crucial for ensuring edits occur only where intended. By forecasting these interactions at scale, PAMmla enables the engineering of new Cas9 variants with improved on-target precision and reduced risk.

“A major outcome of this work is the creation of this PAMmla model that can now be used by researchers to predict customized enzymes that are uniquely tuned for their specific use cases,”
 

Rachel A. Silverstein, PhD Candidate and Study Lead Author, Mass General Brigham

Silverstein, also an NSERC Postgraduate Scholar and 2024 Albert J. Ryan Fellow, led experiments that tested PAMmla-predicted enzymes in models of retinitis pigmentosa, a degenerative eye disease. These enzymes showed enhanced editing specificity, validating the approach.

A Scalable, Open-Access Resource for Gene Editing

Previous attempts at Cas9 protein engineering have been time-consuming and limited in scope. PAMmla changes that by providing a scalable and accessible solution for enzyme customization. The research team has also developed a web-based tool that allows scientists to explore and apply the PAMmla model in their own work.

This tool gives researchers access to a vast collection of custom CRISPR-Cas9 enzymes, dramatically expanding the possibilities for both basic research and therapeutic development.

As genome editing technologies mature, models like PAMmla represent a critical advance, enabling a new era of highly targeted, safer gene therapies.