A comprehensive analysis of 40 blind protein structure predictions

BMC Structural Biology

Table 2 Results of our comparative modelling and fold recognition predictions made at CASP4.

Rating	Difficulty	Target	Fraction of Residues¹	% of Residues¹	RMSD¹ (Å)	Number of Residues²	RMSD² (Å)	Models considered	% sequence identity
E	1	T0128/sodm	202/211	96	1.4	198	1.0	1–5/5	50
E	1	T0122/trpa	235/241	98	2.1	241	2.9	1–5/5	33
E	1	T0123/lacp	145/160	91	3.0	160	4.0	1–5/5	60
E	1	T0099/xxxx	49/56	88	3.0	56	4.7	1–4/4	53
E	1	T0111/eno	425/430	99	1.3	430	1.7	1–5/5	51
E	2	T0125/sp18	125/137	91	3.7	137	4.4	1–4/5	24
E	2	T0113/hcd2	231/251	92	2.0	251	4.4	1–5/5	33
E	3	T0121/malk	228/372	61	3.0	245	3.9	1–5/5	27
G	3	T0112/dhso	290/348	83	3.1	348	4.9	1–5/5	24
G	3	T0103/picp	162/368	44	3.6	156	6.0	1–3/5	26
G	3	T0092/yeco	108/227	48	3.5	104	5.6	1–5/5	12
U	3	T0117/dnk	124/250	63	3.9			1–5/5	21
U	4	T0109/orn	82/182	45	4.3	67	6.3	4–5/5	16
U	4	T0100/pmea	98/342	29	4.0	65	6.0	1/5	10
F	4	T0095/ctn1	33/244	14	4.5			1–5/5	20
U	4	T0127/bchi	95/332	29	3.6	60	5.8	1,3/5	23
U	4	T0101/pell	72/400	18	3.3	74	6.0	1–5/5	22
U	5	T0090/yqie	96/209	48	3.6	107	6.0	1–5/5	19
G	5	T0089/ftsa	124/378	33	3.6	81	5.9	1–5/5	20
F	5	T0108/cbd17	26/179	15	3.4	25	6.1	1–5/5	22
F	5	T0107/cbd9	27/188	14	4.8	28	6.0	1–5/5	19
F	5	T0115/khse	46/296	16	4.5	40	6.0	4–5/5	20
G	6	T0096/fadr	70/222	30	3.6	83	6.0	1–3/5	21
U	6	T0104/yjee	55/158	55	4.0			1–5/5	21
U	6	T0087/ppx1	50/309	16	4.3	54	6.0	1–5/5	17
F	6	T0094/cpdas	35/177	20	3.4	29	6.0	1–5/5	20
U	6	T0120/xrcc4	82/203	40	3.0	96	5.2	5/5	12
F	6	T0116/muts	46/811	6	4.2	50	6.3	1–5/5	11
U	8	T0124/plcb	54/120	22	3.1	60	3.6	2/5	19

The targets (column 3) are sorted by their difficulty (column 2) as provided by the CASP4 assessors (determined by the degree of similarity of the target protein to proteins with known structures). Shown also is a subjective evaluation of the quality of the model (column 1; E – excellent, G – good, U – useful, F – failure), the number of residues (over the total) evaluated using the criterion¹ provided by the CASP4 assessors which considers non-consecutive C_α atoms in the calculation of the RMSD, the percentage of residues evaluated by criterion¹, and the corresponding RMSD¹; the number of residues evaluated using our criterion² which considers only consecutive regions and the corresponding C_α RMSD² (some values are missing using this criteria since the experimental result was not provided to the predictors); the models (out of a total of five) for which the evaluation/result applies; and the percent sequence identity for the alignment between the target and the closest template structure used to construct the model. The data indicate that the modelling performs best on targets with alignments that have > 25% sequence identity to the closest template structure, resulting in 23/29 useful, good, or excellent predictions.

ISSN: 1472-6807