Bremer support

[Derek Sikes]

What a lot of people don't realize when using PAUP is that searches
can spend (waste) large amounts of time on suboptimal islands/
plateaus. This is why Rice et al.'s 1997 search using the rbcL 500
dataset took 13 months. This problem can be avoided by using methods
such as Nixon's Ratchet in NONA  (or use PAUPRat for PAUP), but even
without the Ratchet PAUP can go much faster with the following search
strategy:

1. limit the number of trees swapped during each replicate to avoid
swapping a suboptimal island to completion. This can be done in
different ways. for example by setting in the branch swapping options
(Mac version) the following: click the box to 'save no more than x
trees > score y (each rep)' and set x to a number like 10 or 5 and y
equal to either a known shortest score or simply 1. PAUP will then
burn through all the reps very quickly but doing only quick searches
in each rep. (for non-mac versions see the nchuck command in the
reference guide).

2. When done with step 1 use the best trees found as the starting
trees for an unconstrained search that will explore all swapping
options (swap each island to completion).

The agony, if one watches the search progress closely, is seeing the
search swapping for many days (or weeks) on, say a tree island with
length 1287 and when it is done with that rep and begins the next it
finds a tree island of length 1260 within minutes! This means all
those days spent exploring that 1287 island were wasted. Use the
above strategy to prevent this.

Yours,
-Derek

On 30-Nov-05, at 1:11 PM, Karl Magnacca wrote:

> I should clarify: it was NONA (which is also claimed to be much faster
> than PAUP; it may be used as the regular search engine for TNT but
> I'm not
> sure) that ran so much slower.  The search that failed to find the
> shortest trees in TNT was the special search technique for massive
> data
> sets.
>
>> Well, I suppose different data sets will run differently, but in my
>> original search for MP trees, an approximately 30-minute search in
>> TNT
>> yielded the same set of MP trees as the search in PAUP (500 random
>> addition sequences, TBR branch swapping) which took *almost 2
>> weeks* for a
>> data set with 68 taxa and 120 morphological characters.
>
> What kind of computer are you using?  Unless it's a VIC-20, I can't
> imagine it possibly taking that long for PAUP to run unless something
> isn't working right.  The data set I was working with was 97 taxa,
> ~600
> informative DNA and 15 morphological characters, and it only took 10
> minutes for 100 replicates on a 266 Mhz computer.  On a new one it
> takes
> less than 2 minutes.
>
> Karl

++++++++++++++++++++++++++++++++++++++++++++++
Derek S. Sikes, Assistant Professor
Division of Zoology
Department of Biological Sciences
University of Calgary
2500 University Drive NW
Calgary, Alberta, Canada, T2N 1N4

dsikes at ucalgary.ca
http://homepages.ucalgary.ca/~dsikes/sikes_lab.htm

phone: 403-210-9819
FAX:  403-289-9311

"Remember that Truth alone is the matter you are in Search after; and
if you have been mistaken, let no Vanity reduce you to persist in
your mistake." Henry Baker, London, 1785

Entomological Society of Alberta:
http://www.biology.ualberta.ca/courses.hp/esa/esa.htm
++++++++++++++++++++++++++++++++++++++++++++++