ContactSeek is an AI framework that aims to make genome editing far more specific by focusing on molecular interactions rather than overall structure. In base editing, unwanted edits at off-target DNA sites remain a central obstacle, limiting both research reliability and therapeutic safety. Existing approaches often confront activity–specificity trade-offs and require extensive screening, with low success rates.
The new method, reported in Nature, leverages AlphaFold3’s predicted contact probabilities to detect how DNA and guide RNA interact differently in on-target versus off-target complexes. Rather than relying primarily on predicted three-dimensional conformations, the researchers found that contact probability is more sensitive to the interaction changes that correlate with off-target behavior.
To demonstrate the approach, ContactSeek was applied to Cas9–TadA adenine base editors. The team mapped genome-wide off-targets for these editors and fed the resulting off-target DNA sequences into AlphaFold3 to generate contact probability outputs. By comparing on- and off-target predictions, they identified “consensus contact regions”—clusters of Cas residues showing consistent contact changes with DNA and guide RNA.
From these regions, ContactSeek pinpointed specificity-determining residues, highlighting which amino-acid positions most strongly influence where editing occurs. The framework was designed to be modular: it can be extended to other editors and used to identify key residues in the TadA8e deaminase as well as within Cas protein domains.
The authors report that targeted amplicon sequencing, genome-wide profiling, R-loop assays, and RNA sequencing together confirm markedly improved specificity. Their best engineered variant, combining two mutations in Cas9 and TadA8e, outperformed multiple previously published high-fidelity adenine base editors.
Finally, ContactSeek was generalized to Cas12a-based cytosine base editors, suggesting the strategy is not limited to one enzyme family. Overall, the work proposes an AF3-driven paradigm that integrates structural predictions with interaction-level modeling to guide precision improvements in genome editing tools.
Subject of Research: Precise genome (DNA) base editing specificity; AI-driven contact modelling using AlphaFold3.
Article Title: Precise DNA base editing using AlphaFold3-based contact modelling.
Article References: Meng, H., Lei, Z., Yan, Y. et al. Precise DNA base editing using AlphaFold3-based contact modelling. Nature (2026). https://doi.org/10.1038/s41586-026-10794-z
Image Credits: AI Generated
DOI: https://doi.org/10.1038/s41586-026-10794-z
Keywords: ContactSeek; AlphaFold3; contact probability modelling; genome editing specificity; base editing; Cas9–TadA; Cas12a; off-target prediction; R-loop assay
