DNdisorder: predicting protein disorder using boosting and deep networks

BMC Bioinformatics

Table 1 Performance on the CASP9 dataset

Predictor	ACC		Sensitivity		Specificity		F-measure		Sw		AUC
	Value	±SE	Value	±SE	Value	±SE	Value	±SE	Value	±SE	Value	±SE
cspritz_server	76.00	0.54	66.21	1.0	85.79	0.36	43.46	0.89	52.00	1.1	0.8397	0.005
PRDOS2(291)	75.40	0.61	60.78	1.3	90.03	0.35	47.13	0.88	50.80	1.2	0.8544	0.005
espritz_nopsi_X	74.89	1.2	61.85	1.7	87.93	0.66	44.26	1.9	49.77	2.3	0.8301	0.005
DNdisorder	74.80	0.56	59.70	1.1	89.89	0.21	46.24	0.87	49.59	1.1	0.8299	0.005
PreDNdisorder	74.39	0.58	57.89	1.2	90.90	0.21	46.97	0.90	48.80	1.2	0.8396	0.005
PreDisorder	73.48	0.60	65.47	1.1	81.49	0.64	37.48	0.95	46.96	1.2	0.8136	0.005
biomine_dr_pdb (351)	74.12	1.1	59.45	1.6	88.49	0.67	43.94	1.9	48.23	2.2	0.8205	0.005
Multicom(490)	68.9	0.59	41.34	1.1	95.86	0.23	45.92	1.1	37.8	1.2	0.8550	0.005
DisoPred3C(15)	67.05	1.0	34.90	2.0	99.2	0.07	48.96	2.0	34.11	2.1	0.8539	0.005
iupred_short	63.36	0.69	32.06	0.14	94.67	0.17	34.84	1.3	26.73	1.4	0.6489	0.006

Results of a benchmark of DNdisorder, PreDNdisorder and a number of other disorder predictors. The evaluation was performed on 117 CASP9 targets consisting of 23656 ordered residues and 2427 disordered residues. The standard error (SE) is shown for each performance measure. Not included in this is assessment are 252 residues which were marked as ‘X’ by the CASP assessors. This table also contains four of the top performing methods from CASP9 according to the official CASP9 assessment. The predictions for these methods were downloaded from the official CASP website and the group number for these methods is provided in parenthesis. All values except AUC have been scaled by a factor of 100.

ISSN: 1471-2105