Integrated web service for improving alignment quality based on segments comparison

BMC Bioinformatics

Table 1 General performance of classical methods for building alignments together with segment alignment algorithm incorporating different local structure diversities.

			CE	SEA_T	SEA_F	SEA_I	SEA_loc	BLAST	ALIGN	FFAS
Family (409 pairs)	shift	average		0.61	0.62	0.56	0.49	0.44	0.48	0.49
		>0.9		73	84	69	47	51	60	43
		>0.7		207	231	199	152	146	165	161
		>0.5		282	277	260	215	197	228	227
	RMS	≤ 3.0	257	95	82	82	63	77	54	40
		≤ 5.0	397	237	204	184	147	157	138	118
		≤ 8.0	408	294	269	248	231	196	206	194
	all		409	345	409	404	366	232	372	409
	len		1	0.84	1.14	1.08	0.87	0.56	0.99	1.18

Family-level benchmark for SEA algorithm using FRAGlib's prediction of LSSs (SEA_F) is compared with SEA_I (SEA algorithm using I-sites library), SEA_T, SEA_loc (local single predicted structures), and other classical tools: CE, BLAST, ALIGN and FFAS. The 'average' is the shift score averaged over all the alignments of the whole subset. The numbers of protein pairs with a shift score or RMSD larger than a certain cut-off value in the subset are listed in columns for each program. The counting based on RMSD requires the length of the alignment to be longer than half of its corresponding structural alignment. The 'all' stands for all the alignments with alignment length no shorter than half of the structural alignments. We use the CE for building reference alignments for shift score calculation, as an example of purely structural alignment tool. The 'len' stands for the average alignment length (predicted aligned position / aligned position in reference alignment from CE). We can see that our method provides very long alignments with relatively good overall score. The difference in the values between SEA_T and SEA_F is explained by different lengths of these alignments.

ISSN: 1471-2105