Abstract
CRISPR/Cas9 technologies are now routinely used in plant research, with guide RNA (gRNA) design being a critical determinant of genome editing success. However, rational design of highly active gRNAs is challenging due to complex sequence and biochemical factors affecting activity. While numerous computational prediction tools have been developed, they are predominantly trained on animal cell or microbial data and their performance in plants remains controversial or untested. In this study, using two independent Nicotiana benthamiana experimental datasets comprising a total of 52 gRNAs, we systematically evaluated over 20 freely accessible, Web-based in silico tools for predicting gRNA on-target efficiency. We identified several machine learning-based tools that showed strong correlation with experimental editing efficiency across both datasets. Importantly, gRNAs in the top quartile by prediction score produced significantly higher InDel frequencies than those in the lowest quartile for all tools tested. Furthermore, several algorithms available through CRISPOR, a platform containing a large number of non-model plant genomes, also showed good predictive performance. This may enable better integration of on-target and off-target predictions in gRNA design. Our findings provide practical guidance for improving gRNA design in plant genome editing applications.