Fig. 1
From: Prediction of 8-state protein secondary structures by a novel deep learning architecture
a CRRNN overall architecture. b A local block comprising of two 1D convolutional networks with 100 kernels, and the concatenation (Concat) of their outputs with the original input data. c the BGRU block. The concatenation of input from the previous layer and before the previous layer is fed to the 1D convolutional filter. After reducing the dimensionality, the 500-dimensional data is transferred to the next BGRU layer