Fig. 5

CRISPR/Cas9 deep learning architecture. Artificial intelligence-based deep learning model architecture showing different steps for predicting on/off-targets in the CRISPR/Cas9 system. The model takes a 4 × 23 code matrix corresponding to 4 nucleotides of 23 sequence length as input. The input is passed to the convolutional layer for obtaining sgRNA-DNA matching information by applying different filters of varied sizes. The information is passed for batch normalization to reduce the effect of internal covariates. A pooling layer is connected to the normalization layer which filters out the non-informative values. The result of pooling layer is converted into a single vector by flattening which is connected to the fully connected layer for final model classification