search

beast-dev / beast-mcmc

Bayesian Evolutionary Analysis Sampling Trees
http://beast.community
GNU Lesser General Public License v2.1
181 stars 70 forks source link

birth-death sampling model seems to not work smoothly #645

Closed msuchard closed 8 years ago

msuchard commented 8 years ago

Originally reported on Google Code with ID 637

Hi Walter,

as indicated in the previous email, the birth-death sampling model seems to not work
smoothly in Beast 1, several people reported a problem to me.
A issue I realized when looking into the code carefully right now:

if (!hasFinalSample && n != 1) {
            throw new RuntimeException(
                    "For sampling-through-time model there must be exactly one tip
at time zero.");
        }

The equations are all fine for n>1 and hasFinalSample=FALSE.
We should only allow hasFinalSample=FALSE and fix p=0 (and allow n=1 or n>1, it doesn't
matter).
hasFinalSample=TRUE might be useful for phylogenies with fossils, but we didn't test
it properly.

Did we ever compare Denise's birth-death skyline model in Beast 2 (for 0 rate shifts)
to the birth-death model in beast 1? they *should* be the same...

Cheers Tanja

Reported by dong.w.xie on 2012-06-25 02:12:32

msuchard commented 8 years ago 
Hi Walter,
there seems to be some problem in Beast 1, see below?
Could you doublecheck and get back?
Thanks a lot, Tanja

Begin forwarded message:

From: Emmanuelle MULLER <emmanuelle.muller@cirad.fr<mailto:emmanuelle.muller@cirad.fr>>
Date: June 18, 2012 2:15:01 PM GMT+02:00
To: <tanja.stadler@env.ethz.ch<mailto:tanja.stadler@env.ethz.ch>>
Subject: a question about your model in BEAST

Dear Tanja,
I was at your seminar at Montpellier University last friday and was very interested
by the models you have implemented in BEAST.
I am working on epidemiological data concerning Cacao swollen shoot virus.
Last friday, I have tested your model with my sequences (the joined file containig
sequences from 4 genetic groups collected from 1993 to 2010) but it doesn't work.
I used  your "epidemiological birth and death" model implemented in BEAST and I obtain
a message I also joined to you  because it is  not enough clear for me.
Could you help me?
Thank you in advance,
Emmanuelle Muller

Sequence data is in email.

Reported by dong.w.xie on 2012-06-25 02:14:36

msuchard commented 8 years ago

Reported by rambaut on 2012-08-01 16:16:18

msuchard commented 8 years ago 
change to:

        if (!hasFinalSample && n < 1) {
            throw new RuntimeException(
                    "For sampling-through-time model there must be at least one tip
at time zero.");
        }

Reported by dong.w.xie on 2012-08-10 00:54:55