Multiple alignment accuracy.

Quality score statistics were measured in the 60 sequence sets of the IRM-1 database [96]. “Percent perfect” refers to the proportion of the 60 datasets in which all homologous residues were correctly aligned. All programs were run with default parameters, except that Program 1 and DIALIGN-TX used the parameters detailed in Table 3. Because all programs other than Program 1 produce global multiple alignments as a matter of course, the quality score credits them for aligned residues independently of whether these residues are identified as lying within a conserved region. None of these programs explicitly identifies such regions, although DIALIGN-TX does so implicitly, as described in the caption to Table 3. Accordingly, in order not to artificially handicap Program 1 on this test, we calculated its quality scores by aligning, immediately adjacent to the conserved pattern it identifies within each sequence, and without gaps, all the residues deemed to lie beyond this pattern. In the small fraction of cases where the identified pattern stops short of the boundary of the embedded motif (see Table 3), this can produce a slightly better quality score than the pattern, considered in isolation, would yield. CPU execution times are for programs run on a Dual Pentium 4 Xeon 3.0 GHz CPU Linux computer with 64-bit architecture, and are averaged over three runs.