beast mrca and offset value

[sehrish kakakhel]

Dear all,

i am trying to find divergence time for my plant data. I use GTR model default clock rate. i set mean  36.45  and Sd 13 and off set 27. when i decrease off set value it effect the tree ages. and also when i increase mcmc runs the tree ages increase up to 250 mya . so which offset value be best to get correct result. and how should we know that our result are accurate. kindly guide.

[Alexei Drummond]

Is it a normal or a log-normal distribution? You shouldn’t use an offset for a normal distribution, since the mean is already a location parameter. But I wouldn’t use a normal distribution at all because it provides support for negative ages and this can produce weird interactions with the tree prior in general. You should use a log-normal distribution since it only has support for positive ages and doesn’t exhibit the kind of behaviour you are describing. Ages are scale parameters, so you should use a prior that is defined only for positive values (like log-normal).

If you are already using a log-normal distribution then you should be aware that the SD is in *log* units, and 13 would be a very extreme level of prior uncertainty.

On 22/11/2023, at 9:11 PM, sehrish kakakhel <sayyar...@gmail.com> wrote:

Dear all,

i am trying to find divergence time for my plant data. I use GTR model default clock rate. i set mean  36.45  and Sd 13 and off set 27. when i decrease off set value it effect the tree ages. and also when i increase mcmc runs the tree ages increase up to 250 mya . so which offset value be best to get correct result. and how should we know that our result are<offet-28-IM runs.jpeg><offset 30-20M runs.jpeg> accurate. kindly guide.

--
You received this message because you are subscribed to the Google Groups "beast-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beast-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/4c2d1ed3-5810-48dc-9232-c9a8b1637454n%40googlegroups.com.
<offet-28-IM runs.jpeg><offset 30-20M runs.jpeg>

[sehrish kakakhel]

THANKS Alexei drummond. I have used log normal distribution in add prior. I am attaching a screenshot. Are all my parameters correct?

please guide me

To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/F26A6797-89CD-4BF8-A281-98923545EAF0%40gmail.com.

[Alexei Drummond]

M=36 and S=13 parameters are both very extreme. These parameters represent the mean and standard deviation of the *log* of the node age. The median of your prior as specified is 6.76 trillion as you can see from the plot on the righthand side of your screenshot. 

If you mean to have a mean=36.45 and a standard deviation around 13 you could have parameter values M=3.536, S=0.346 which gives a median of 34.33 and a 95% range of 17.42 to 67.64:

> qlnorm(p=c(0.025, 0.5, 0.975), meanlog=3.536, sdlog=0.346)

[1] 17.42407 34.32933 67.63648

I usually use the qlnorm function in R to find parameters that give me a median and 95% HPD that I am happy with. But you can also just change the parameters in BEAuti until the plot on the right represents the distribution that you want. If it is easier for you then you can check the “mean in real space” checkbox and then you can directly set M parameter to 36.45, but the S parameter will still be the standard deviation of the *log* of the prior age, so in your case it should be around 0.3 to 0.35, not 13.

Cheers

Alexei

To view this discussion on the web visit https://groups.google.com/d/msgid/beast-users/CAHaUptLxM44qCfwzD8GfO%3Dr8E-DGJqBirNuqtuDkuK1f-wTsow%40mail.gmail.com.
<priors.jpeg>