Fig. 4From: Predicting subcellular location of protein with evolution information and sequence-based deep learningBox chart of prediction accuracy on benchmark datasets D3106 and D4802. The three plots in left column are box charts of average precision (AP), ranking loss (RL), coverage (COV) on dataset D3106, and the plots in right column are box charts of corresponding results on dataset D4802. Model BLSTMÂ +Â ConvNet1Â +Â ConvNet2 is the best one among the four models with highest average precision and smallest divergence on dataset D4802. However, the ranking loss and coverage of this model is greater than the ones of model BLSTMÂ +Â ConvNet1. On dataset D3106, model BLSTMÂ +Â ConvNet1Â +Â ConvNet2 is better than the other three models in average precision, ranking loss and coverage. In average precision and coverage, those four models perform better on dataset D3106 than on dataset D4802. The ranking loss values of those four models on dataset D4802 are lower than the values on dataset D3106. a This plot shows the average precisions of model BLSTM, model BLSTMÂ +Â ConvNet1, model ConvNet2 and model BLSTMÂ +Â ConvNet1Â +Â ConvNet2 when they are tested on dataset D3106 with five-fold cross validation. Model BLSTMÂ +Â ConvNet1Â +Â ConvNet2 has the greatest average precision than the other three models, while model BLSTMÂ +Â ConvNet1 has stable performance. b The average precisions of those four models when they are tested on dataset D4802. Model BLSTMÂ +Â ConvNet1 has the best performance on this dataset, however model BLSTMÂ +Â ConvNet1Â +Â ConvNet2 is more stable on this dataset. c The ranking loss values of those four models when they are tested on dataset D3106. d The ranking loss of them with dataset D4802. e The coverage results of predictions by model BLSTM, model BLSTMÂ +Â ConvNet1, model ConvNet2 and model BLSTMÂ +Â ConvNet1Â +Â ConvNet2 on dataset D3106. f The coverage results of predictions by those four models on dataset D4802Back to article page