Beast 1.10.1 and Speciation:Birth-Death model error

[jeanbapt...@googlemail.com]

Dears Beast Creator(s),

I am working on Herpesvirus on Oyster with 23 samples with whole genome sequence. The sampling have a wide time-span from 20008 to 2017 (usefull for estimates nodes dates) and from various host species (from various part of the world).
I have three questions (newbies regarding phylogeny) :

1- Does a wrong/bad demographic models will affect nodes date estimates ?

2- What is the most robust demographic model with sparse sampling regarding time and space ?

3- I tried various coalescent models with success, but when I launch a "Speciation: Birth-Death Serially Sampled" (BDSS) I am a error message :

"
Sep 11, 2018 9:10:50 AM dr.app.beast.BeastMain <init>
SEVERE: Parsing error - poorly formed BEAST file, All_samples_out_human_vivaldi_without_abalone.aligned.beast.v8.xml:
Error parsing '<scaleOperator>' element with id, 'null':
Scale operator can only be used on parameters with an infinite upper or lower bound (use a RandomWalk) (bdss.relativeDeathRate)

Error thrown at: dr.inferencexml.operators.ScaleOperatorParser.parseXMLObject(Unknown Source)
"
I tried to modify by hand the beast xml without success. I have attached the beast xml with the (BDSS) model...
I attached the beast xml with bdss model to reproduce the error.



Thank you for maintaining this help groups and I will be happy with any though or remarks (even: Read the F$*k?ng manual part XXX)
Cheers,

JB
PS: I posted, by error, an incomplete version of this message previouly... You could delete it without any problem.

[Guy Baele]

Hi JB,

This is indeed a bug; you can replace the scaleOperator in the XML by the following:

<randomWalkOperator windowSize="0.75" weight="3" boundaryCondition="logit">

   <parameter idref="birthDeath.relativeDeathRate"/>

</randomWalkOperator>

Please let me know if this works.

Best regards,
Guy

Op dinsdag 11 september 2018 09:52:58 UTC+2 schreef jeanbapt...@googlemail.com:

[Sebastien Calvignac-Spencer]

Hey Guy,

I ran into the same kind of problem when running speciation models (birth death and bd incomplete sampling). I replaced the scale operators but then got the following error message:

SEVERE: Parsing error - poorly formed BEAST file, 180917_VP1_1.xml:
Error parsing '<randomWalkOperator>' element with id, 'null':
The logit transformed RandomWalkOperator cannot be used on a parameter without bounds.

Error thrown at: dr.inferencexml.operators.RandomWalkOperatorParser.parseXMLObject(Unknown Source)

The according parametres (birthDeath.meanGrowthRate and  birthDeath.relativeDeathRate) however definitely have finite bounds in my xml (the default ones uniform 0-100000 and 0-1).

Thanks a lot for your help!

Seb

[Guy Baele]

The logit transformed RandomWalkOperator should only be used for parameters with a [0,1] domain.

The prior choices for the other parameters are arguably quite poor (I know some of these are the default choices in BEAUti), i.e. uniform with an arbitrarily high upper bound.
I would suggest to change these to a lognormal distribution and estimate them using a scaleOperator.

I will try to modify the default prior choices for the next BEAST release.

Best regards,
Guy

Op maandag 17 september 2018 09:41:48 UTC+2 schreef Sebastien Calvignac-Spencer:

[Kamil F.]

Hello there, 

I have similar problem. Trying to run dating with Beast - when I use Yule process all is fine, however when I switch to preferred Birth-death process (the rest remains unchanged) I get exactly same error as previously mentioned. 

I edited xml file switching scaleOpeartor for randomWalk, however then what I get are results with no much biological sense (I'm dating phylogeny of relatively young group, more or less 15-20 milion years old). With Yule process the root is estimated at ca. 16 Myo, when switched to Birth-death it changes for over 200 Myo (which makes absolutely no sense). Simultaneously, setting a prior on root height makes it go all the way throughout the analysis from over 350 Myo to 0 Myo...

I'd be much grateful for any help. 

I'm attaching the original (unedited) xml file and what beast returns.

Best wishes, 

Kamil