Table 3 Performance comparison of the Met-predictor with other existing methods on the independent test set I.

Residue	Methods	MCC	ACC	SEN	SPE	PRE	CPRE	AUC	PRAUC
K	MEMO	0.104	0.528	0.106	0.950	0.679	0.075	—	—
	MASA	0.164	0.531	0.067	0.994	0.923	0.301	—	—
	PLMLA	0.061	0.531	0.517	0.544	0.531	0.042	—	—
	PmeS	0.209	0.542	0.083	1.000	1.000	1.000	—	—
	MethK	0.101	0.558	0.019	1.000	1.000	1.000	—	—
	iLM_2L	0.075	0.531	0.239	0.822	0.573	0.049	—	—
	MONO	0.039	0.450	0.218	0.814	0.649	0.648	—	—
	DI	0.151	0.522	0.044	1.000	1.000	1.000	—	—
	TRI	0.291	0.578	0.156	1.000	1.000	1.000	—	—
	GPS-MSP	0.130	0.517	0.033	1.000	1.000	1.000	—	—
	MONO	−0.094	0.383	0.000	0.986	0.000	0.000	—	—
	DI	0.106	0.511	0.022	1.000	1.000	1.000	—	—
	TRI	Nan	0.500	0.000	1.000	Nan	Nan	—	—
	Met-predictor(seq)	0.195	0.597	0.633	0.561	0.591	0.053	0.611	0.606
	MONO	0.126	0.617	0.845	0.257	0.641	0.641	0.594	0.705
	DI	0.223	0.611	0.578	0.644	0.619	0.351	0.557	0.553
	TRI	0.194	0.594	0.469	0.719	0.625	0.265	0.611	0.549
	Met-predictor(seq + str)	0.261	0.631	0.644	0.617	0.627	0.061	0.655	0.647
	MONO	0.136	0.622	0.864	0.243	0.642	0.642	0.587	0.699
	DI	0.291	0.644	0.578	0.711	0.667	0.400	0.660	0.601
	TRI	0.221	0.609	0.531	0.688	0.630	0.269	0.664	0.585
R	MEMO	0.282	0.624	0.386	0.862	0.736	0.104	—	—
	MASA	0.316	0.622	0.305	0.939	0.833	0.172	—	—
	PmeS	0.176	0.587	0.498	0.675	0.605	0.060	—	—
	GPS-MSP	0.192	0.550	0.122	0.977	0.844	0.181	—	—
	MONO	0.113	0.293	0.048	1.000	1.000	1.000	—	—
	DI	0.024	0.505	0.049	0.961	0.556	0.384	—	—
	Met-predictor(seq)	0.352	0.675	0.637	0.714	0.690	0.085	0.723	0.731
	MONO	0.097	0.746	1.000	0.013	0.745	0.745	0.541	0.755
	DI	0.073	0.633	0.223	0.837	0.404	0.404	0.583	0.579
	Met-predictor(seq + str)	0.355	0.677	0.630	0.723	0.695	0.086	0.734	0.746
	MONO	0.126	0.746	0.978	0.075	0.753	0.753	0.574	0.778
	DI	0.122	0.662	0.194	0.894	0.476	0.475	0.628	0.617

Two versions of Met-predictor are included here: Met-predictor(seq), where only sequence-based features are used to build models, and Met-predictor(seq + str), which uses not only sequence-based features, but also the novel structure model-based features to build models. The lysine residue is represented by “K” while the arginine corresponds to “R”. The definitions of measures MCC, ACC, SEN, SPE, PRE, CPRE, AUC and PRAUC are shown in “Performance Evaluation” section and Eqs. (1) to (6). The “Nan” for the MCC, PRE and CPRE is because both TP and FP are zero, resulting in the measure divided by zero. There are no AUC and PRAUC values for other methods because other tools or servers do not output the probabilities for their predicted sites or only gives the probabilities of positive sites they predicted. Note that the probabilities of both positive and negative samples are needed to calculate the AUC and PRAUC values.

Search