Researchers have used DeepMind’s AlphaFold3 to create ContactSeek, an AI‑driven framework that predicts contact probabilities between base‑editing proteins and DNA. By identifying key residues that interact most aggressively with DNA, the team introduced just two mutations—one in the Cas9 domain and one in TadA8e—to produce adenine base editors with dramatically higher precision and activity while retaining editing efficiency. The approach was also adapted to LbCas12a‑based cytosine editors, demonstrating its modularity. All data and code have been released publicly.
Off‑target effects remain the biggest safety hurdle for gene‑therapy applications. By systematically improving enzyme specificity through AI‑guided protein redesign, ContactSeek could shorten development timelines, lower regulatory concerns, and attract biotech investment focused on safer gene‑editing therapeutics.