Table 11. Timing comparison of FSA and other methods on 16S sequences.

Program	100	200	300	400	500 seqs
ClustalW	1,194 s	4,147 s	9,110 s	16,187 s	27,755 s
DIALIGN	4,346 s	19,449 s	49,388 s	(fail)	(fail)
FSA –fast	1,513 s	3754 s	5,641 s	9,767 s	15,683 s
FSA –fast –noindel2 –refinement 0	638 s	1,495 s	2,467 s	3,604 s	5,154 s
MAFFT	31 s	105 s	243 s	442 s	54 s
MUSCLE	351 s	1,235 s	1,516 s	4,384 s	7,552 s
ProbConsRNA	16,319 s	(fail)	(fail)	(fail)	(fail)
T-Coffee	1,362 s	3,666 s	7,880 s	15,254 s	22,085 s
SeqAn::T-Coffee	3,024 s	(fail)	(fail)	(fail)	(fail)

Comparison of runtimes of FSA and other alignment methods when aligning 16S ribosomal sequences. MAFFT was faster than any other method by an order of magnitude; the next-fastest programs were MUSCLE and FSA. FSA can be made substantially faster by using a 3-state, rather than the default 5-state, HMM (with little loss of accuracy; see Table 8) and disabling iterative refinement. MAFFT was run with the –auto option, which presumably triggered a faster alignment mode on the 500 sequence dataset than was used for the datasets with fewer sequences. The designation “(fail)” means that a programs failed to align a dataset (generally due to out-of-memory errors). Timing results are from computers with 2.40 GHz CPUs and 2 GB of RAM. 16S sequences were obtained as a random slice of prokMSA from Greengenes [65] and had an average length of 1,450 nt.