Skip to main content
Fig. 3 | BMC Bioinformatics

Fig. 3

From: Predicting subcellular location of protein with evolution information and sequence-based deep learning

Fig. 3

ROC curves (Receiver operating characteristic curve) of model BLSTM, BLSTM + ConvNet1, ConvNet2 and BLSTM + ConvNet1 + ConvNet2, on benchmark dataset D4802. This figure plots one fold validation result. Detailed information about the fivefold cross validation results can be found at Additional file 1: Table S4 and Figure S2. Those plots shows that those four models perform worse on dataset D4802 than on dataset D3106. a 33 curves of model BLSTM. The best AUC is 0.9121, however the worst one is 0.7072. The average AUC value of this model is 0.7696. The plot shows that a small group of curves are centered together, while other curves are divergent from one another. b 33 curves of model BLSTM + ConvNet1. Two curves around the diagonal show that the accuracy of prediction on two subcellular positions is bad. The average AUC value of this model is 0.8543 which is better than the average value of model BLSTM. c ROC curves of model ConvNet2. This plot is similar to the (c) in Figure2 but with several curves under diagonal. So the prediction on some subcellular locations isn’t better than random guess. The max AUC value is 0.8986 and the min value is 0.5080. The average AUC value is 0.6806. d In this plot, there is still one curve that is under diagonal. However, the best AUC value reaches to 0.9434 and the average AUC value reaches to 08594. In general this model performs better than the other three models

Back to article page