Leveraging Generative AI To Design Novel, Functional Deaminases For Adenine Base Editing

By Matt Nethery, Principal Data Scientist, Computational Biology

Generative artificial intelligence and machine learning models offer a powerful path to transcend the evolutionary limits of natural protein sequence space. Training computational models on massive datasets of carefully filtered and structural-validated sequences enables the de novo creation of novel adenine deaminases. Selecting optimal sampling parameters balances generative creativity with structural integrity, producing diverse enzymes with low sequence identity to known wildtype variants. Iterative machine learning optimization rapidly improves baseline activity, driving adenine-to-guanine base editing performance from initial low single digits to over 20% on-target conversion. Crucially, these engineered candidates display exceptional specificity, yielding minimal bystander editing even at adjacent nucleotide positions.

Download the poster presentation to see how computational enzyme creation unlocks precise, tailor-made gene editing tools for previously intractable therapeutic targets.