Improving pairwise sequence alignment accuracy using near-optimal protein sequence alignments

BMC Bioinformatics

Table 1 Sequence and structural alignment scores for two example alignments.

	Levitt-Gerstein Structural Similarity	Sequence Alignment Score^†	Shift Score vs. Dali	Levitt-Gerstein Structural Similarity	Sequence Alignment Score^†	Shift Score vs. Dali
Alignment	1 hdaB00 vs. 1 mytA00 (pair 25)			1bcgA00 vs. 1b7dA00 (pair 42)
CE	2333	85	0.986	978	61	0.920
Matras	2330	103	0.983	966	49	-
DALI	2334	115	1.000	980	24	1.000
LSQMAN	2337	-20	0.946	980	24	0.940
Optimal Sequence^§	2352	145	0.940	583	108	0.837
95% Neighborhood^§	2355	145	0.900	696	104	0.880
75% Neighborhood^§	2359	145	0.923	812	88	0.780
probA^§	2362	37	0.645	830	37	0.757
robustness^¶	2301	-526	0.734	383	24	0.565
model^¶	2352	145	0.941	579	108	0.796

^†Semi-global alignment, Blosum50 substitution matrix, gap open penalty -10, gap extension penalty -2.
^§ The sequence with the highest Levitt-Gerstein similarity score was chosen from the set of optimal alignments, the set of alignments with a sequence similarity score within 95% of optimal, the set of alignments with a sequence similarity score within 75% of optimal, and the set of probA alignments, and the sequence similarity score and shift score for that alignment were recorded.
^¶Alignments were created using the log-odds score produced by a logistic regression model in place of a substitution matrix. "Robustness" refers to a model that had only robustness as an independent variable, while "model" refers to the full model (incorporating robustness, edge frequency, and maximum bits-per-position). See text for details.

ISSN: 1471-2105