Residue propensities, discrimination and binding site prediction of adenine and guanine phosphates

BMC Biochemistry

Table 4 Cross-prediction prediction AUC (in%) of trained models, measured by using model trained for binding sites of one ligand to predict binding sites of the other ligands

Tested	AMP	ADP	ATP	GMP	GDP	GTP	CAMP	PCG
Trained
AMP	100.00	73.27	74.89	72.83	75.11	76.30	75.37	62.88
ADP	81.12	100.00	86.22	72.81	80.34	77.64	73.64	74.48
ATP	85.34	86.38	100.00	73.94	78.16	78.63	73.23	74.70
GMP	73.13	70.02	71.14	100.00	74.21	71.24	75.00	61.32
GDP	75.57	75.58	73.94	85.16	100.00	80.93	70.28	70.56
GTP	74.45	74.29	74.09	67.32	83.31	100.00	66.69	59.67
CAMP	70.17	69.77	70.15	66.96	69.19	70.92	100.00	91.20
PCG	64.86	65.00	64.64	62.56	66.09	65.61	77.82	100.00

Please note that all data of a given category was trained, allowing to over-learn on itself and hence self-prediction values (trained and test data being the same) for all models showed 100% correct predictions. This is in contrast to cross-validation performance results in Table 3.

ISSN: 1471-2091