Divergence and Clock rate in multilocus analyses

[Guillermo Friis]

Dear BEAST2 users,

I'm sorry a lot of things  but I couldn´t find solutions from other sources, meaning bibliography, BEAST2 book and of course this group. I'll order my questions following the usual workflow to generate the xml file using BEAUTi.

First of all, I have the next partitioned dataset for 85 samples of the songbird Junco spanning several divergent populations from all the three current species:

-ND2: with a accepted divergence rate of ~2% for birds.

-Dloop.

-Other mitochondrial markers.

-A set of nuclear markers.

No calibrations available.

1. First doubt I had was about the so-called meaning of the substitution rate in 'Site Model' label. It looks that it's useful when you have partitions to compute relative rates. I ran BEAST with estimate checked but fixing the mean substitution rate. Never converged. To what point could be problematic to leave it at 1.0 and don't estimate it for all partitions?

2. Clock Model: how can I do to use the standard rate of 0.02 for ND2 to constraint the rest? Should I use 0.02 or the half of it since it's a divergence rate and not a substitution rate? It's better to use a strict clock since it's a standard value, or it's better to use the lognormal uncorrelated distribution to account for bias from this rate? Any advice in the 'Number of discretes Rates' prior to set? What's exactly the Normalize option?

Thanks everybody

[Remco Bouckaert]

Hi Gulliermo,

On 3/04/2015, at 4:18 am, Guillermo Friis <guill...@gmail.com> wrote:

> 1. First doubt I had was about the so-called meaning of the substitution rate in 'Site Model' label. It looks that it's useful when you have partitions to compute relative rates. I ran BEAST with estimate checked but fixing the mean substitution rate. Never converged. To what point could be problematic to leave it at 1.0 and don't estimate it for all partitions?
>

The substitution rate in the site model panel is multiplied with the clock rate to get a rate for each branch. If there is no strong prior on the substitution rate, then clock and substitution rate are unidentifiable, so that would explain the lack of convergence.

If you estimate them and select Fix mean substation rate, then substitution rates can be interpreted as relative rates, and then analysis has some chance to converge.

In BEAUti v2.2.1 this is the default whenever a substitution rate is estimated.

> 2. Clock Model: how can I do to use the standard rate of 0.02 for ND2 to constraint the rest? Should I use 0.02 or the half of it since it's a divergence rate and not a substitution rate? It's better to use a strict clock since it's a standard value, or it's better to use the lognormal uncorrelated distribution to account for bias from this rate? Any advice in the 'Number of discretes Rates' prior to set? What's exactly the Normalize option?
>

Many of these answers can be found in the BEAST book, available from http://beast2.org/book.html.

Hope this helps,

Remco