Biology Reference
In-Depth Information
computationally more expensive. Similarly, the use of Opal to align
subsets will require a reduction in subproblem size because of Opal's
memory requirements (and hence increased running time).
Less is known about how to set the maximum subproblem size
when ClustalW is used for aligning subsets, but the default settings
are probably fine.
Decomposition edge
. This algorithmic parameter determines how the
dataset is decomposed—through the centroid edge (which pro-
duces a roughly equal decomposition into two datasets) or the
longest branch. The default is the centroid edge, and this produces
results of similar accuracy to the longest edge, while being much
faster.
Stopping rule
. There are various settings that determine the stop-
ping condition and when it is evaluated. You can set the stopping
rule to be defined by the number of iterations or time, and at least
one of these must be specified (selecting both means that either can
trigger the stopping rule). You can begin this stopping rule imme-
diately (“After Launch”) or only after the ML score stops improv-
ing (“After Last Improvement”). We recommend using “After Last
Improvement” unless your dataset is so large that you need to limit
the number of iterations. “Blind mode” means that the previous
iteration's tree will always be used as the tree in the beginning of the
next iteration, regardless of its ML score. Disabling “blind mode”
means that the best-scoring tree so far will be used in the beginning
of the next iteration, which can cause the iterative search to become
stuck in local optima. We recommend enabling “blind mode”.
Final tree/alignment.
This determines which tree is the output for
the SAT´ run. The “Best” setting returns the best-scoring align-
ment/tree pair encountered during the SAT
´
analysis. The “Final”
setting returns the final alignment/tree pair from SAT´.
CPUs available
. This is the number of CPUs in your machine that
SAT
´
should use, and using multiple CPUs can speed up the
analysis. However, do not set this number to more CPUs than
you have!
See Note 4
.
Extra RAxML search
. Checking this box makes SAT´ perform a
RAxML analysis of the final alignment. If time is not highly con-
strained (and your dataset is not too large), checking this box is
recommended if you have used FastTree for the ML tree estimator.
However, when you use RAxML for the ML tree estimator,
it automatically computes a RAxML tree on the final alignment,
and so it does not make any sense to check this box.
Two-phase (not
SAT´
)
. Check this box to run a two-phase analysis
(first align and then compute an ML tree) using the settings in the
“External Tools” window, instead of running a SAT´ analysis. This
may not produce alignments and trees as accurate as those pro-
duced by SAT´, but should be faster.
