Fig. 5. Development of deep learning-based prediction models, collectively named DeepSniper.

a, A simplified schematic representation of DeepSniper development. b,c, Performance of DeepSniper in predicting the activities of Sniper1 and Sniper2L at matched (b) and mismatched (c) target sequences using target sequences that were not included in training datasets. The Pearson’s correlation coefficients (r) and the Spearman’s correlation coefficients (R) are presented. The number of target sequences (n) is n = 5,100 and 5,069 for Sniper1 and Sniper2L, respectively (b) and n = 295 for both Sniper1 and Sniper2L (c).