Comparison of predicted prime editing efficiencies (DeepPrime score) from a deep-learning-based model 44 trained on editing results in HEK293T cells using an optimized prime editor (PEmax) and hMLH1dn to observed prime editing (PE) efficiencies in mouse embryos microinjected at the two-cell stage with PE4 components (editor mRNA, pegRNA, mMLH1dn mRNA) from this study. Each dot represents the specific pegRNA design used in our study. Color shade indicates the relative predicted score of the pegRNA compared to the maximum score predicted by the DeepPrime-FT model across all feasible pegRNA designs ( Methods ) for a given target site/edit. Pearson correlation coefficient ( r = 0.68) reported with P value ( P = 0.01) from two-sided t -test. Dashed line represents fit from linear least-squares regression. Throughout our study, asterisks specify use of the optimized PEmax editor (PE2*, PE4* methods), as opposed to the PE2 editor 21 (PE2, PE4 methods). Data in a f are compiled from multiple experiments (Supplementary Tables 7 9 and Methods ). For c and d , lowercase letters indicate edit and black lines connect results for the same edit across conditions. For c e , P values are from two-sided Student’s t -tests. For box plots, boxes indicate the median and interquartile range (IQR) for each group of embryos with whiskers extending 1.5 IQR past the upper and lower quartiles.
