. 2006 Aug 26;34(16):4364–4374. doi: 10.1093/nar/gkl514

Table 2.

Average alignment scores in tests of multiple sequence alignment programs

Methods/Models	Testing datasets
	SCOP 0–10%	SCOP 10–15%	SCOP 15–20%	SCOP 20–40%	SCOP	PREFAB	SABmark	SABmark	BAliBASE3.0
	(355)	(432)	(420)	(578)	All (1785)	(1682)	Sup (425)	Twi (209)	Q/col (218)
HMM_1_1_0	0.313	0.514^a	0.727^a	0.885	0.644^a	0.723^a	0.516^a	0.193^a	0.862/0.551
HMM_1_1_1	0.313	0.512^a	0.728^a	0.886	0.644^a	0.724^a	0.512^a	0.186^a	0.861/0.550
HMM_3_1_1	0.321^a	0.514^a	0.730^a	0.888	0.647^a	0.726^a	0.516^a	0.186^a	0.862/0.554
HMM_1_3_1	0.327^a	0.518^a	0.732^a	0.889^a	0.650^a	0.729^a	0.519^a	0.194^a	0.863/0.554
HMM_3_3_1	0.329^a	0.520^a	0.733^a	0.889^a	0.651^a	0.731^a	0.522^a	0.196^a	0.863/0.557
ProbCons	0.291	0.486	0.702	0.879	0.625	0.716	0.485	0.166	0.862^c/0.556^b
MAFFT-fftnsi	0.283	0.472	0.673	0.865	0.608	0.7	0.45	0.147	0.829^c/0.515^c
MAFFT-einsi	0.293	0.498	0.71	0.882	0.631	0.72	0.502	0.175	0.866^c/0.585^b
MAFFT-linsi	0.301	0.5	0.707	0.883	0.633	0.722	0.51	0.184	0.868^c/0.586^b
MAFFT-ginsi	0.308	0.497	0.714	0.888	0.637	0.715	0.495	0.176	0.840^c/0.526^c
MUSCLE	0.262	0.453	0.662	0.866	0.597	0.68	0.433	0.136	0.816^c/0.472^c
ClustalW	0.21	0.357	0.566	0.798	0.519	0.617	0.39	0.127	0.749^c/0.373^c

The format of the HMM names (‘HMM_solv_ss_u’) is explained in Table 1. Average Q-scores are shown for all the testing datasets. For the BAliBASE3.0 dataset, both the Q-score (‘Q’, first number) and column score (‘col’, second number, fraction of entirely correct columns) are shown. The first four testing datasets are representative SCOP40 domain pairs with added homologs. SABmark has ‘superfamily’ dataset (sup) and ‘twilight zone’ dataset (twi). The number of alignments in each testing dataset is shown in parentheses and the identity range in % is specified above the number of alignments for SCOP datasets. MUMMALS implementing different HMMs are the first five methods. All sequences pairs are subject to consistency measure in MUMMALS. The best scores of MUMMALS and the best scores of other programs are in bold.

^aMUMMALS with this model is statistically better than the best of other programs according to Wilcoxon signed-rank test (P < 0.015).

^bFor BAliBASE3.0 test, the difference between MUMMALS with model HMM_1_3_1 or HMM_3_3_1 and this program is not statistically significant (P > 0.05) according to Wilcoxon signed-ranks test.

^cFor BAliBASE3.0 test, MUMMALS with model HMM_1_3_1 or HMM_3_3_1 is statistically better than this program (P-value less than 0.01, except for Q-scores of ProbCons, for which P = 0.017).