Login

Kale · 05-02-2024, 06:03 PM

As far as I can tell yes f2_from_precomp is effectively qpFstats with a different name. As such it is also very fast like qpFstats.
I have to use an abbreviated dataset to prepare the F2 data (which is the commend extract_f2), not to use it (f2_from_precomp), again just like I have to with qpFstats on admixtools1.
I don't think F2_from_geno will solve anything. If anything it might slow it down if the name reflects how it works.

Jerome · (This post was last modified: 05-03-2024, 05:15 AM by Jerome.)

(05-02-2024, 06:03 PM)Kale Wrote: As far as I can tell yes f2_from_precomp is effectively qpFstats with a different name. As such it is also very fast like qpFstats.
I have to use an abbreviated dataset to prepare the F2 data (which is the commend extract_f2), not to use it (f2_from_precomp), again just like I have to with qpFstats on admixtools1.
I don't think F2_from_geno will solve anything. If anything it might slow it down if the name reflects how it works.

Thanks!
One last thing, I have read that sometimes f2_from_precomp can give different results and it's preferable to run directly with geno file in prefix for Qpadm in admixtools2?
Is this true?

Also,what function do we have to call get gendstats through f2_from_precomp?
PrintF/Fatalx?

Or with the details=Yes in paramater file?(I don't think it's needed in AT2)

Kale · 05-05-2024, 12:43 AM

That is more to do with extract_f2. When you generate f2 stats there are two major settings to consider, maxmiss and afprod.
Maxmiss=0 uses only the snps covered by all populations that you are generating F2 for. Note that if you generate F2 for a bunch of pops, and you decide only to use a subset of them in your test, you still only get the snps that were covered in the total bunch. That being said, maxmiss=0 is probably most suited for 'rotating' qpadm, or qpgraph, where you are going to be using the same populations for a large number of trials. I haven't played around with afprod settings much with maxmiss=0, but I think generally the default is afprod=false.
Maxmiss=1 is like setting allsnps=TRUE. For whatever reason, maxmiss=1 + afprod=false is a big-nono. The results in many cases are fine, but there are some that flat out contradict direct from genotype data. If maxmiss=1 then must afprod=TRUE.

The results from precomputed F2 do vary slightly from direct from geno, but it's not anything too crazy. Here was a test I did in 2021 when first started using admixtools2.

Show Content

I don't remember how to get gendstats, details=YES sounds right.

kolompar · Yesterday, 08:12 PM

(04-29-2024, 02:04 PM)Kale Wrote:
(04-29-2024, 08:35 AM)Jerome Wrote: Thanks!
The results are interesting,onge is still present,in greater quantities than before but I think this is a case of Wezmeh in particular who seems to be pretty different compared to other iran_n samples like the GanjDareh_N.
Could you model that sample similarly?

Also,now mota/mbuti/deep ancestry takes a major hit,and it seems it's not needed anymore In that large amounts,this does cast a doubt on the basalness of iranN,if only 6% mota/deep Ancestry is needed.

Zlaty Kun is a bit more of an early crown Eurasian so it may eating up a bit of deep ancestry, I wonder what the results will be if bacho Kiro IUP is used instead of Zlaty Kun.
Ust ishim could work too but he has a bit more eastern affinity.

A minor request,could you try this model with BachoKiro_IUP instead of Zlaty Kun,and MA1 instead of AG3?(since MA1 is more eastern than AG3 and the ANE in iranN could be more AG3 and using AG3 may be inflating the Onge).
Also,GanjDareh instead of Wezmeh as the target so the results could be compared a bit to lazaridis' 2018 modelling with Dzudzuana.
Iran_N's ANE is more related to AG3 than to MA1, and I would argue that AG3/MA1 are about equally East/West, if there is any Eastern leaning, it would be from AG3 due to a trifle of UKY/Kolyma stuff.

Iran_GanjDareh_N
BachoKiro_IUP 0.232202 0.128585 1.80582
Kostenki14 0.275729 0.0478508 5.76226
AG3 0.191370 0.0177896 10.7574
Andaman_100BP.SG 0.198286 0.0799078 2.48144
Mota.SG 0.102413 0.0158679 6.45410
Tail: 0.07
right = c('SouthAfrica_2000BP.SG', 'Ust_Ishim.DG', 'Papuan.DG', 'China_UP', 'RUS_Primorsky_DevilsCave_N.SG', 'GoyetQ116_1', 'Muierii1', 'BachoKiro_BK1653', 'Yana_UP.SG', 'MA1.SG', 'Peru_RioUncallane_1800BP.SG')

It is a haaardddd fail if ZlatyKun is put in the right.

Maybe I didn't get your pops right but it passed for me with Zlaty Kun in the right. Only had 63k SNPs though and there were a bunch of warnings about singularity.
But more importantly Iberomaurusian in the right breaks it and Natufian doesn't fix it completely and Mota still remains too. Proof for Basal, finally?

Kale · 11 hours ago

IBM in the right breaking it is unsurprising, just shared Western stuff between them and Iran_N. Natufian can't fix it because any sort of ANA will substantially drop ZlatyKun affinity.

Login
Username/Email:
Password:	Lost Password?
	Remember me