Estimating divergence times with substitution rates

[pseu...@gmail.com]

Hi,

I am a new user to BEAST 2.0 and want to estimate divergence dates based on "known" substitution rates of a locus. However, I am very confused regarding how I should set up my analysis in BEAUTI. 

I've read the online version of the BEAST 2 book and saw that the substitution rate of a locus equals its mutation rate times the clock rate. I used to think these 2 were equivalent (i.e. if you know the cytb diverges about 2% every million years that IS the clock rate). Could somebody please explain this to me?

Regarding the practical part of the analysis: 

-Basically I want to estimate divergence dates for my dataset, which consists of sequences for the cytb and a nuclear intron. I have an estimate of the substitution rate for the cytb obtained from the literature (average= 1.5% range= 0.5-2.5%). However, there are no estimates for the other marker. So my main question is how should I proceed with my analysis?

1) I know I have to link the trees (i.e. a concatenated analysis since I am not using *Beast), but am dubious whether to link the clocks or not.

2) I am giving the substitution rate for the cytb, (including the upper and lower bounds) under the 'Site Model', and I assume I have to check the 'estimate' box for those bounds to really be incorporated in the analysis? I am also clicking the 'estimate' box for the other locus since it is not known. I am then 'fixing the mean substitution rates' since I understand thats the only way a clock can be calculated? but I am not sure if this makes any sense.

Any insights you might have will be GREATLY appreciated.

Many thanks beforehand!

Mariano

[Remco Bouckaert]

Hi Mariano,

Typically, BEAST uses substitution rates for relative rates, and clock rate for absolute rates. But in the end, the actual rate used is the product of these two. BEAUti just makes it convenient to add a prior on substitution rates (by checking the Fix substitution rate check box) such that the mean rate of the substitution rates is one.

> 1) I know I have to link the trees (i.e. a concatenated analysis since I am not using *Beast), but am dubious whether to link the clocks or not.

In general, if you have good reason to believe these rates are the same, you can link them, but if not, then you should leave them unlinked. Given that you have a mitochondrial and nuclear sequence, I would leave them unlinked.

2) I am giving the substitution rate for the cytb, (including the upper and lower bounds) under the 'Site Model', and I assume I have to check the 'estimate' box for those bounds to really be incorporated in the analysis? I am also clicking the 'estimate' box for the other locus since it is not known. I am then 'fixing the mean substitution rates' since I understand thats the only way a clock can be calculated? but I am not sure if this makes any sense.

It is correct that to set up a prior using the known rates from the literature, you need to estimate the parameter. However, since you are using absolute rates, instead of using the rate in the Site model panel, I would use the clock rate in the Clock Model panel. In any case, do not use fix the mean substitution rate, since you are interested in absolute rates, not relative ones. If you fix the mean rate and put a prior on the cytb that is very low, the one on the intro will become large to ensure the mean rate remains 1.

Note that to estimate the clock rate (if you have not other calibrations) you need to uncheck the 'Mode/Automatic set clock rate'.

Cheers,

Remco


 

[pseu...@gmail.com]

Dear Remco,

Thanks a lot for taking the time to clearly explain this. It was insanely useful.

I might be wrong, but I have the feeling that this distinction between absolute and relative rates (especially to which panel they correspond to in the program) is not crystal clear to many using it.

Thanks a lot again,

Mariano