Fig. 3From: Predicting protein-ligand binding residues with deep convolutional neural networksArchitecture of the deep convolutional neural network in en-DeepCSeqSite (enDCS-SI). The encoder of enDCS-SI is the same as that of stdDCS-SI. The decoder of enDCS-SI is designed to extract the information form the labels or the previous prediction. The decoder of stdDCS-SI is included in the decoder of enDCS-SI, where the weights of the former are fine-tuned during training enDCS-SI. ‘p’, ‘s’ and ‘e’ represent padding, start mark and end markBack to article page