PHYLIP DNA likelihood programs for the Cray YMP

Ron DeBry debry at ds1.scri.fsu.edu
Wed Feb 6 15:04:14 EST 1991


If anyone is interested in running Joe Felsenstein's DNA maximum
likelihood phylogeny program (dnaml) on a Cray under UNICOS, I have
a version which is much faster (on a Cray, that is) than the
distribution version.  I replaced the innermost loops of two of the most
frequently called subroutines with vectorizable FORTRAN modules, which
are linked to the rest of the program with segldr.  It works here on a
YMP 4/432 - I guess that it should also work as is on an XMP - I have no
idea whether it will work properly on a Cray-2.

As a rough benchmark, I ran a 15 species, 2000 bp data set, with global
branch swapping, in less than 15 minutes of cpu time.  Your mileage will
vary, depending on the number of patterns and numerous other parameters.

I have also added an option which I needed badly - E for Eliminate taxa.
This allows you to keep the same input file with all the species, but
only analyze a subset.  I have tested this fairly rigorously so far, and
am continuing to try & dream up ways to make it crash before I let
anyone else try it.  

I intend to also modify the dnamlk program (maximum likelihood with an
assumed molecular clock), although I haven't actually done it yet.  If
someone asks for it, I'll be more likely to finish that quickly, since I
don't have a need to run that program right now myself.

I do not intend to modify the restriction site maximum likelihood
program, for two reasons.  First, the innermost loops in that program
are very small (the number of iterations equals the length of the
recognition sequence), so there would only be a small gain in speed, or
even a loss in some cases.  Second, restriction site data sets are
likely to be much smaller than sequence data sets.  I have a nine
species, 140 site data set that runs in less than a minute of cpu time
with the unmodified program.  If you have a really massive restriction
site data set, let me know, and we can discuss what might be done.

If you want the modified dnaml program, contact me at one of the
following addresses (ds1 is preferred):

InterNet:  debry at ds1.scri.fsu.edu   or  debry at fsu.scri.fsu.edu

BitNet:    DEBRY at FSU

Mail:      Ron DeBry
	   Department of Chemistry
	   Florida State University
	   Tallahassee, FL 32301

Phone:     (904) 644-4233




More information about the Bio-soft mailing list