skyline plot interpretation

[Michael Harvey]

Peter,

I am playing around with skyline plots of theta/migration and trying to convert the timescale. In the manual you say: "You can calculate the absolute time by multiplying the scale by generation time times mutation rate per year (per site for DNA; per locus for all other datatypes)." It would seem, however, that doing this would result in very small numbers(?). The timescale on my plots from sequence data analysis is between 0 and 150. If I assume a generation time of 1 year (typical in my study species) and a typical substitution rate/site/yr on the order of 10e-8, the timescale on my plots only goes from 0 to 0.000015 years(!). Did you mean to divide by mutation rate per year? When I do that, the number seems ridiculously large.

Thanks for any suggestions,

Mike Harvey

[jahol...@gmail.com]

You actually need to divide by the mutation rate per generation (i.e., u * generation time). The results are obviously very sensitive to both of these values being correct (or as close to correct as possible). To get more resolution over recent events, you may wish to adjust the binsize smaller. Peter suggests not having too much confidence in values going back further than theta. I've been doing some tests and find that replicate runs are consistent with this - the farther back you go the more noise I get between replicate runs. For my species with a mutation rate similar to yours but a longer generation time the R^2 between replicate runs is pretty good up to about 100k ybp. Going further back they start to diverge wildly.

Jason

[Stênio Foerster]

Hi Jason and all users,

Please, could you be more specific about how to interpret (convert) these time scales? 

For example, If I have a peak in M at time 0.003 (from the x-axis of the histogram printed in the outfile.pdf), how can I interpret this value in terms of absolute time? 

The mutation rate for the analyzed locus is 0.014/myr and the generation time is 3.5 years.

Thanks in advance! 

 

   

Em quarta-feira, 30 de outubro de 2013 11:26:43 UTC-2, jahol...@gmail.com escreveu:

You actually need to divide by the mutation rate per generation (i.e., u * generation time). The results are obviously very sensitive to both of these values being correct (or as close to correct as possible). To get more resolution over recent events, you may wish to adjust the binsize smaller. Peter suggests not having too much confidence in values going back further than theta. I've been doing some tests and find that replicate runs are consistent with this - the farther back you go the more noise I get between replicate runs. For my species with a mutation rate similar to yours but a longer generation time the R^2 between replicate runs is pretty good up to about 100k ybp. Going further back they start to diverge wildly.

.

[Peter Beerli]

I think Jason’s answer is good (also given some time ago)

if you have a time t=0.003 and mu per generation then this would be 

t/mu —> now time in generations, with a generation time of 3.5 years that means divide the result by the generation time

but your mu is per million year and not per generation -> t/mu —> 0.003/( 0.014) = 0.2142857143 million year —> 214,285 years

(I also say that without dated samples the estimates will be not very good).

Peter

--
You received this message because you are subscribed to the Google Groups "migrate-support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to migrate-suppo...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/migrate-support/5bf5ee0c-df7b-4ead-8bb7-3412a0347babo%40googlegroups.com.