Using structural motif descriptors for sequence-based binding site prediction

BMC Bioinformatics

Table 1 Matches for ATP- and ADP-binding descriptors with various E-value thresholds.

Hits	Precision	Recall	Threshold	TP	FP	New
2873	96.59	7.55	1.00E-10	2775	98	771
3113	96.5	8.17	1.00E-09	3004	109	814
3459	96.18	9.05	1.00E-08	3327	132	854
3943	96.25	10.32	1.00E-07	3795	148	899
4488	95.9	11.7	1.00E-06	4304	184	929
5114	93.86	13.05	1.00E-05	4800	314	955
5773	92.74	14.56	0.0001	5354	419	982
6477	91.71	16.15	0.001	5940	537	1012
7477	91.53	18.61	0.01	6844	633	1053
8749	91.45	21.76	0.1	8001	748	1115
10349	89.43	25.17	1	9255	1094	1249
14165	78.86	30.38	10	11171	2994	1499
27926	53.67	40.76	100	14988	12938	2440
93261	25.85	65.56	1000	24109	69152	6397

A benchmark set containing 205.000 known ATP-binding or not ATP-binding SwissProt entries was used to measure the precision and recall of 306 ADP/ATP binding descriptors. Hits were counted at various E-value cutoffs. Number of True Positives (TP), False Positives (FP) and number of those True Positives not found automatically by any Prosite-pattern (New) were determined.

ISSN: 1471-2105