Table 1.

Comparison with other bioinformatics tools.

Bioinformatics tool	Performance (%)
	Train set	Test set¹	Experimental data²
SignalP 4.07	99.61	99.27	51.64
LipoP8	99.83	99.71	61.39
Phobius46	98.78	98.72	72.08
PRED-TAT10	99.66	99.61	62.07
Classifier
Preprotein (#P1)	97.19	95.51	97.96
MatureP (#M22)	91.46	91.24	85.10
Disorder (#M7)	84.73	84.18	85.86

We measured the performance of four bioinformatics tools: SignalP 4.0 LipoP, Phobius and PRED-TAT on the training, testing and experimental datasets. We used the AUC as a performance metric³⁵. AUC depicts relative trade-offs between true positive (benefits) and false positive (costs) and represents the performance of the average classifier (over different classifiers which assume different miss-classification cost ratios).

¹Randomly selected samples (20% of the total sample set; Table S1) which remained unused during the training of the classifiers.

²Experimental data manually collected from the literature (Table S4).