Extended Data Fig. 3: Identification of guide features that influence on-target activity in TP53 tiling screen etoposide arm.

(a) Rule Set 3 Sequence score of TP53 tiling screen positive control guides, namely those that introduce missense, nonsense, or splice-site mutations not previously identified as benign, discretized by activity. The results of this analysis are presented separately with the use of a z-score cutoff of −2, −3, or −4 to define guides as active. Significance is calculated between active and inactive guides using the Mann-Whitney Wilcoxon one-sided test. This analysis incorporates data with the use of both the ABE8e and TadCBEd editor. n = number of guides represented in each activity bin. (b) (a), but with the use of the BE-Hive model to predict base editing guide efficacy instead of Rule Set 3 Sequence score. n = number of guides which fall in each activity bin. (c) (a), but with the use of the FORECasT-BE model to predict base editing guide efficacy with ABE8e. n = number of guides represented in each activity bin. (d) (c), but estimating model performance at predicting base editing guide activity with TadCBEd. n = number of guides represented in each activity bin. (e) Sequence motifs representing nucleotide identities at each position relative to the editable nucleotide that are enriched among active positive control guides. Active guides are defined as those with a z-score below −2, whereas inactive guides represent those with a z-score between −2 and 2. Analysis is stratified by base editor.