Figure 2
From: OutCyte: a novel tool for predicting unconventional protein secretion
The OutCyte-SP model and its predictions. (a) The structure of the convolutional neural network for learning the motifs at the N-termini of sequences. The network consists of two convolutional layers, which use ReLu transformations and no pooling. A max pooling layer follows to extract the strongest distinguishing features, followed by the dense and softmax layers. (b) Matthews correlation coefficients (MCCs) for signal peptide identification from three datasets are shown in the left panel. In the right panel, micro-averaged MCCs were calculated for OutCyte-SP and DeepSig on the two evaluation datasets. *The SignalP5.0 training dataset overlapped with SignalP4.0’s benchmark set; thus, two MCCs were not included. (c) Intersections among 4 different annotations for signal-peptide-containing proteins in the human proteome from OutCyte-SP, UniProt (with evidence), SignalP 4.1 and DeepSig.
