Sequence embedding for fast construction of guide trees for multiple sequence alignment

Table 1 mBed performance on the ten biggest Pfam/HOMSTRAD families.

Name	Size	Len	%ID
				(1)	(2)	(3)	(4)	(1)	(2)	(3)	(4)	(1)	(2)	(3)	(4)
PF01381	9993	53	23	-	25	55	136	764	57	55	175	13.3	26.7	25.3	34.7
PF00006	9796	209	43	-	134	248	280	4364	48	49	88	42.8	36.6	36.6	38.0
PF00989	9681	95	17	-	43	88	197	1281	50	51	159	46.5	33.3	31.8	34.1
PF00486	9615	75	30	-	34	69	107	950	55	52	104	63.9	92.8	64.9	89.7
PF00571	9551	119	19	-	73	143	268	1993	54	50	152	6.15	3.08	1.54	1.54
PF00097	9423	41	33	-	18	38	94	517	44	43	115	53.2	54.8	61.3	54.8
PF01479	9352	47	32	-	17	40	90	496	45	46	124	58.3	91.7	89.6	79.2
PF00046	9305	54	35	-	20	43	85	651	41	42	77	59.4	44.9	46.4	60.9
PF00550	9249	63	25	-	28	59	136	794	47	47	141	51.3	32.9	55.3	59.2
PF00149	9072	198	14	-	133	256	552	3515	47	46	172	75.4	71.9	72.3	76.1
Average	9503	95	27	0	53	104	195	1533	49	48	131	47.0	48.9	48.5	52.8

The ten biggest Pfam entries containing 9,000-10,000 sequences, which have a corresponding HOMSTRAD alignment are used here. Four different methods were applied to each entry to calculate a distance matrix. These methods are: (1) the traditional process of calculating a full distance matrix from the sequence data using an alignment distance measure; (2) mBed default; (3) mBed followed by the 'usePivotObjects' method; (4) mBed followed by the 'usePivotGroups' method. A UPGMA guide tree is constructed from each matrix and used as a guide tree for progressive alignment of the sequences. The alignment is then scored against the corresponding HOMSTRAD structural alignment using Column Score.
(1) Full d(x, y) distance matrix; (2) mBed; (3) mBed + usePivotObjects; (4) mBed + usePivotGroups

ISSN: 1748-7188