Researchers have deployed Google DeepMind’s AlphaFold3 to build ContactSeek, a new computational framework that redesigns CRISPR-Cas9 gene-editing proteins to drastically reduce off-target DNA cutting. Unveiled in a July 22, 2026 study in Nature, the advancement addresses gene therapy’s biggest safety hurdle by improving precision without sacrificing intended activity.
How ContactSeek Uses AlphaFold3 to Predict Mismatches
Gene therapy has long wrestled with a persistent headache: the tendency of molecular editing tools to accidentally cut or alter the wrong stretches of DNA. To combat this, researchers turned to Google DeepMind’s AlphaFold, the Nobel Prize-winning protein-folding software. Updated versions are specifically designed to handle interactions between proteins and nucleic acids, as well as complexes of multiple proteins.
Initially, the research team fed AlphaFold versions of a target DNA sequence along with a guide RNA, the Cas9 sequence, and an enzyme that chemically modifies bases and sticks to Cas9. Undeterred, the team simplified the input by feeding AlphaFold only the DNA, RNA, and the Cas9 protein, since Cas9 is the primary factor determining sequence specificity.
This streamlined approach produced a structure that closely agreed with actual experimental observations of nucleic acids and proteins. By comparing the structures generated when fed different on- and off-target sites, the researchers spotted a clear pattern. While roughly two-thirds of off-target sites caused the Cas9 protein to adopt a slightly different structure, over 95 percent of off-target sites altered which amino acids contacted the RNA.
This meant that in many cases, Cas9 maintains its normal overall shape while internal amino acids flex to accommodate mispaired bases at off-target sites. Capitalizing on AlphaFold’s built-in ability to identify contact probability
—the chance that items like amino acids or nucleotides sit within a tiny distance of eight Angstroms—the team compared contact probability outputs for on- and off-target sites. This pinpointed exact amino acid shifts, leading the team to name their computerized analysis setup ContactSeek.
Engineering Safer Base Editors and Modular Adaptability
ContactSeek initially generated a large list of shifting amino acids. To narrow things down, researchers focused on the specific regions of the Cas9 protein where these amino acids clustered, interpreting the clustering as a sign that those areas were adapting to mismatched bases. They then tested versions of Cas9 carrying different amino acids at those specific sites.
The team focused primarily on adenine base editors built from the Cas9-TadA system.
Their best-engineered variant incorporated just two mutations: one affecting the Cas9 domain and another affecting TadA8e. Furthermore, the method proved modular, as the team successfully adapted ContactSeek for LbCas12a-based cytosine base editors, which represent an entirely different class of editing tool.
Compressing Development Timelines for Therapeutics
Off-target genetic modifications remain the single largest regulatory and safety hurdle holding back the broader approval of gene therapies. By utilizing predictive AI models rather than years of trial-and-error mutagenesis, frameworks like ContactSeek compress the development timeline significantly.
To support further open science and collaborative development, all data and code for ContactSeek v1.0.0 have been made publicly available on Zenodo and GitHub. As the first CRISPR-based therapies continue reaching patients, tools that systematically optimize editing precision point toward a more reliable regulatory pathway for future genomic medicines.
Keep reading
