Skip to main content
. 2021 Nov 25;12(12):1878. doi: 10.3390/genes12121878

Figure 3.

Figure 3

Overview of R-CRISPR. The input on- and off-target sequence pairs is represented into a 7 × length (i.e., length represents the length of gRNA-target sequences) binary matrix as the input of the feature extraction layer. The feature extraction layer contains 40 convolutional kernels and 40 modified RepVGG modules composed of a 1 × 3 convolutional kernel, a 1 × 1 convolutional kernel, and an identity branch. The output of the feature extraction layer is then passed to the bi-directional recurrent layers, each direction is based on 15 LSTM units, to learn sequential patterns of the feature matrix. Followed by the recurrent layer, there are two dense layers with sigmoid as the activation function for final outputs.