Predicting base editing outcomes using position-specific sequence determinants

Nucleotide-level control over DNA sequences is poised to power functional genomics studies and lead to new therapeutics. CRISPR/Cas base editors promise to achieve this ability, but the determinants of their activity remain incompletely understood. We measured base editing frequencies in two human cell lines for two cytosine and two adenine base editors at [~]14,000 target sequences. Base editing activity is sequence-biased, with largest effects from nucleotides flanking the target base, and is correlated with measures of Cas9 guide RNA efficiency. Whether a base is edited depends strongly on the combination of its position in the target and the preceding base, with a preceding thymine in both editor types leading to a wider editing window, while a preceding guanine in cytosine editors and preceding adenine in adenine editors to a narrower one. The impact of features on editing rate depends on the position, with guide RNA efficacy mainly influencing bases around the centre of the window, and sequence biases away from it. We use these observations to train a machine learning model to predict editing activity per position for both adenine and cytosine editors, with accuracy ranging from 0.49 to 0.72 between editors, and with better generalization performance across datasets than existing tools. We demonstrate the usefulness of our model by predicting the efficacy of potential disease mutation correcting guides, and find that most of them suffer from more unwanted editing than corrected outcomes. This work unravels the position-specificity of base editing biases, and provides a solution to account for them, thus allowing more efficient planning of base edits in experimental and therapeutic contexts.