Slowness of MrBayes due to partitioning or missing Nexus block parameters?

[Giuseppe Emanuele De Pasquale]

Hello there!

I'm encountering an issue with the computational time of Mr. Bayes.

I attempted to create two phylogenies using the same input file (an alignment of mitochondrial genes, tRNAs, and the D-loop concatenated). The first was without a Nexus block appended at the end, while the second included a Nexus block.

In the first case, I initiated the analysis using all default parameters, only adjusting a few parameters such as the number of generations and the substitution model.

For the second analysis, I appended the Mr. Bayes Nexus block to the file to utilize substitution models identified with Modelfinder. There were six partitions, each with its own model.

Here's the problem: The first analysis computed 1,000,000 generations in about three and a half hours (approximately 4,700 generations per minute), whereas the second one computed 670,000 in eight hours (approximately 1,400 generations per minute).
I had set 15 million generations, but I stopped the analysis due to its slowness.

I'm seeking to understand if this difference is caused by some missing parameter in the Nexus block or if the partitioning is slowing down the analysis.

Any help is greatly appreciated! Thanks in advance!


To help you better understand, I'll provide you with the parameters for both analyses:


-First analysis (no nexus block)

MrBayes on XSEDE - Parameters
Covarionopts_ false
Nbetacatopts_ 5
Parsmodelopts_ false
aamodelpropts_ fixed(poisson)
allchainsval_ allchains=No
atocrate_ 1.0
atograte_ 1.0
atotrate_ 1.0
betameanx_ 1.0
betavary_ 1.0
brlensprexp1_ 10.0
brlenspropts_ unconstrained:exponential
burninfracval_ 0.25
codingopts_ all
covswitchpropts_ uniform
covswitchuni1_ 0.0
covswitchuni2_ 100.0
ctograte_ 1.0
ctotrate_ 1.0
flagdatatype_ dna
gtotrate_ 1.0
mcmcdiagnval_ mcmcdiagn=Yes
minpartfreqval_ 0.1
mrbayesblockquery_ false
nchainsval_ 4
ngenval_ 1000000
nocharsets_ 0
nrunsval_ 2
nstopts_ 6
nswapsval_ 1
nucmodelopts_ 4by4
ordertaxaval_ Ordertaxa=Yes
pinvarpropts_ uniform
pinvarpruni1_ 0.0
pinvarpruni2_ 1.0
precision_ 15
ratecorrpropts_ uniform
ratecorrpruni1_ -1.0
ratecorrpruni2_ 1.0
rateopts_ equal
ratepropts_ fixed
relburninval_ relburnin=Yes
reportancstateopts_ false
reportsiterateopts_ false
revmatopts_ dirichlet
revmatpropts_ dirichlet
run_version_ 7
runtime_ 5
samplefreqval_ 1000
sbrlensval_ Savebrlens=Yes
scientific_ false
set_beagle_params_ true
shapeprdir2_ 50.0
shapepropts_ uniform
shapepruni1_ 0.0
specify_diagnfreqval_ 5000
statewfreqprdir1_ 1.0
statewfreqpropts_ dirichlet
stopruleval_ stoprule=Yes
stopval_ 0.01
sump_burninfrac_ 0.25
sump_relburnin_ Yes
sumpburnin_ 10
sumpnruns_ 2
sumt_burninfrac_ 0.25
sumt_conformat_ Figtree
sumt_relburnin_ Yes
sumtburnin_ 10
sumtcontype_ contype=Halfcompat
sumtdisplaygeq_ 0.05
sumtnruns_ 2
sumtntrees_ 1
sumtshowtreeprobs_ showtreeprobs=No
swapfreqval_ 1
symdirihyperpropts_ fixed(infinity)
tempval_ 0.200
topologypropts_ uniform
tratioopts_ beta

-Second analysis (with nexus block)

BEGIN mrbayes;

charset part1 = 1-952\3 1996-2335\3 3712-4003\3 4006-5821\3 6349-7888\3 7891-8569\3 8572-9349\3 9352-10486\3;
charset part2 = 2-953\3 956-1994\3 1997-2336\3 2339-3710\3 3713-4004\3 5825-6347\3 6350-7889\3 7892-8570\3 8573-9350\3 9353-10487\3 10490-11165\3 11168-11366\3;
charset part3 = 3-954\3 957-1995\3 1998-2337\3 2340-3711\3 3714-4005\3 4008-5823\3 6351-7890\3 7893-8571\3 8574-9351\3 9354-10488\3 10491-11166\3 11169-11367\3;
charset part4 = 955-1993\3 2338-3709\3 4007-5822\3 10489-11164\3 11167-11365\3 12337-13289 13290-14872 14873-16230;
charset part5 = 5824-6346\3 5826-6348\3;
charset part6 = 11368-12336;
partition my_partition = 6: part1, part2, part3, part4, part5, part6;
set partition = my_partition;

lset applyto=(1) nst=2 rates=gamma ngammacat=4;
prset applyto=(1)
statefreq=fixed(0.293775,0.233708,0.229822,0.242696)
shapepr=fixed(0.507914)
revmat=fixed(1,10.58276562,1,1,10.58276562,1);

lset applyto=(2) nst=2 rates=gamma ngammacat=4;
prset applyto=(2)
statefreq=fixed(0.196152,0.258776,0.123297,0.421775)
shapepr=fixed(0.52895)
revmat=fixed(1,4.049911056,1,1,4.049911056,1);

lset applyto=(3) nst=6 rates=gamma ngammacat=4;
prset applyto=(3)
statefreq=fixed(0.388858,0.302098,0.0637054,0.245338)
shapepr=fixed(1.23511)
revmat=fixed(0.4825695163,26.05426455,0.7430936236,1.255519089,14.83038161,1);

lset applyto=(4) nst=2 rates=invgamma ngammacat=4;
prset applyto=(4)
statefreq=fixed(0.344283,0.226782,0.158858,0.270077)
pinvar=fixed(0.424522)
shapepr=fixed(1.00963)
revmat=fixed(1,8.920532676,1,1,8.920532676,1);

lset applyto=(5) nst=2 rates=gamma ngammacat=4;
prset applyto=(5)
statefreq=fixed(0.238088,0.0653413,0.321794,0.374777)
shapepr=fixed(0.277141)
revmat=fixed(1,31.68811338,1,1,31.68811338,1);

lset applyto=(6) nst=2 rates=invgamma ngammacat=4;
prset applyto=(6)
statefreq=fixed(0.284944,0.256525,0.14199,0.316541)
pinvar=fixed(0.406094)
shapepr=fixed(0.432569)
revmat=fixed(1,10.49438137,1,1,10.49438137,1);

mcmcp

nchains=4
ngen=15000000
samplefreq=1000
savebrlens=yes
printfreq=1000

checkpoint=yes
checkfreq=1000

append=no;

mcmc;

sumt burnin=0.20;
sump burnin=0.20;

log stop;
END;

[Mark Miller]

Hi all, my colleague, Wayne Pfeiffer, manages efficiency here at CIPRES.

He has this response:

Hi Giuseppe,

 

I help maintain the CIPRES gateway.

 

Partitioning the multiple-sequence alignment typically slows down phylogenetic analysis because it reduces the effectiveness of parallelization. The slowdown that you observed is not surprising.

 

Best regards,

Wayne