Figure 6.

Relationship between sgRNA sequence features and activity. (A, B) Relative importances of features (Gini importance) that contribute to the prediction power of the gradient boosting regression tree for the (A) Cas9 and (B) eSpCas9 model. The features of each model are sorted and the most important 20 features are shown. (C, D) Coefficients of features that contribute to the prediction power of the L1 regularized linear regression for the (C) Cas9 and (D) eSpCas9 model. The features for adjacent dimers (order2_Px_xx) are plotted as heatmap according to their relative position (‘Px’) in sgRNA. Positive coefficients suggest positive contributions to sgRNA activity and vice versa. For all plots shown here, results are shown from models trained on all the available data (training plus test set in Figure 5A).