dealing with fixed derived alleles
Andi Kautt
I have a question regarding the impact of fixed derived alleles in the DSFS.
What I mean by that is sites that are fixed in the focal populations as compared to an outgroup
(populating the uppermost right field in a 2D-DSFS).
I was under the assumption that, since these sites are actually monomorphic in the focal populations,
they would not affect the parameter estimates (or only marginally). But in my case it seems that removing
these sites, i.e. setting the uppermost right field in the DSFS to zero, changes the inferred parameter values
quite a lot.
I understand that monomorphic sites are used in a combination with a mutation rate to gain actual demographic
units, but the number of derived fixed SNPs in comparison to the monomorphic ancestral sites (lowermost left corner)
is minute.
To give a more practical example: I built a multidimensional SFS for two focal populations and the spectrum was
polarized with an outgroup. In total ca. 2 M sites were screened and there are ca. 3000 polymorphic sites between
the two focal populations. Further, ca. 9000 sites are differently fixed between the two populations and the outgroup.
So the DSFS will have a bit less than 2 M in the entry for monomorphic sites (lowermost left field), a sum of 3000 distributed throughout the
spectrum and an entry of 9000 in the uppermost right corner. But simulations with these 9000 fixed derived sites and such without,
but otherwise with exactly the same settings, lead to consistently different parameter estimates. Without these sites the estimates
are actually quite close to analyses using the folded SFS.
But my question is: Do such sites actually contain useful information or should they be removed?
Any advice would be much appreciated.
Sorry for this long post and I hope I made myself clear.
Cheers,
Andi
Laurent Excoffier
note that for fastsimcoal2, a derived allele has necessarily occurred since the MRCA of all sampled lineages. Therefore unless there have been multiple mutations at a given site, there cannot be any fixed derived alleles.
So if you have fixed sites, they should have mutated above the MRCA. Thus I would simply put them into the monomorphic category (0,0).
This category is important for getting absolute values of your parameters, and having different numbers of monomorphic sites will affect the parameters.
best
laurent
Bas T
Dear Laurent,
I have a follow-up question on using fixed-derived alleles. I am using a 3-population setup with three pairwise joint SFSs. When I use the best estimated parameters of my model fitted including fixed-derived alleles to calculate expected SFSs with OUTEXP, the expected SFS still contains a non-zero fully derived entry. I also noticed that if I set the fully derived category in the observed SFS to zero, the likelihood changes dramatically.
I was thinking that fastsimcoal2 might ignore fully derived alleles because it cannot estimate the divergence to the outgroup. As you explained previously, all derived mutations should be more recent than the MRCA of the sampled ingroup.
Therefore, I do not understand why OUTEXP outputs a non-zero fully derived category. Even with the infinite-sites model (-I), the expected SFS still contains a non-zero fully derived entry. Is this because an allele can theoretically be fixed derived in two of the three populations, but not necessarily in all three?
How should I handle the fully derived category? Should I exclude all sites that are fixed derived in all three populations and treat them as monomorphic when computing the three pairwise joint SFSs?
Best regards,
Bastiaan