Application of protein structure alignments to iterated hidden Markov model protocols for structure prediction

BMC Bioinformatics

Table 3 Examples of possible underpredictions detected in the SCOP database.

Probe Super-family	Model Super-family	Probe Example	Model Example [Built From Domain(s)]	Model Type	*hmmpfam* E-value	CE Z-score	Total # of Hits
c.3.1	c.4.1	d1lpfa2	d1lqta2 et al. (except m1gtea4)	SLAHMM	1.3 × 10^-5	4.1	8
c.30.1	c.4.1	d1kjqa2	d1djna3	Single-master	0.1	4.2	4
c.3.1	c.30.1	d1m6ia2	d1iow_1 et al. (except d1ehia1)	SLAHMM	0.12	3.7	8
c.66.1	c.3.1	d1inla_	d3grs_2 et al. (except d1gpea1)	SLAHMM	0.82	4.2	2 (PS2)
c.66.1	c.30.1	d1jgla_	d1gsoa2	Single-master	0.87	4.4	2 (PS4)
c.78.2	c.30.1	d1b74a1	d1gsoa2 et al. (except d1ehia1)	SLAHMM	0.0045	4.1	2

These class-level matches were treated as fold-level matches for purposes of the experiment because of their clear structural and sequence similarity. Although only a one way match is shown in the example cases provided, matches were bi-directional between superfamilies with an E-value of less than 1 in all cases (see text for details). In the case of the single-master HMM examples, the domain used to generate the HMM is provided. For the SLAHMM examples, only the domain used in the CE alignment is provided, although these alignments contain all the domains from the superfamily except one (which is mentioned). In both cases, the CE Z-score reported is for the probe domain aligned to the domain provided with the model. The reported hmmpfam E-value is that provided by comparison of the probe with the model. The total number of hits refers to the number of bi-directional hits with E < 1 observed between the given superfamily pair after all possible probe and model combinations. The number provided is from the results of searches with PS1, unless otherwise noted (two relationships were not detected with PS1, but were detected using results from other parameter sets).

ISSN: 1471-2105