Fig. 1

Overview of DeepEnhancer. A raw DNA sequence is first encoded into a binary matrix. Kernels of the first convolutional layer scan for motifs on the input matrix by the convolution operation. Subsequent Max-pooling layer and batch normalization layer are used for dimension reduction and convergence acceleration. Additional convolutional layers will model the interaction between motifs in previous layers and obtain high-level features. Fully-connected layers with dropout will perform nonlinear transformations and finally predict the response variable through softmax layer