related recombination problem In sex chromosome test

[AprilYUZhang]

Currently, I test sex chromosome with recombination
I set up recombination to happen to "m" individuals at the middle point; see the following block.

m 0 0 0 1 1 1
m 1 1 1 0 0 0

Interestingly, there are two problems, which might help us improve recombination estimate.

Here is my genotype file

fm 1 1 1 1 1 1
ff 0 0 0 0 0 0
mm 0 0 0 0 0 0
mf 1 1 1 1 1 1
f 1 1 1 1 1 1
m 1 1 1 1 1 1
pfem 1 1 1 2 2 2
pmal1 0 0 0 1 1 1
pmal2 1 1 1 0 0 0

pedigree file

fm 0 0 1 # 0 for male, 1 for female
ff 0 0 0
mm 0 0 1
mf 0 0 0
f ff fm 0
m mf mm 1
pfem f m 1
pmal1 f m 0
pmal2 f m 0

expected_hap.txt

f 9 9 9 9 9 9 # Paternal haplotype
f 1 1 1 1 1 1 # Maternal haplotype
ff 9 9 9 9 9 9
ff 0 0 0 0 0 0
fm 1 1 1 1 1 1
fm 0 0 0 0 0 0
m 0 0 0 1 1 1
m 1 1 1 0 0 0
mf 9 9 9 9 9 9
mf 1 1 1 1 1 1
mm 0 0 0 0 0 0
mm 0 0 0 0 0 0
pfem 1 1 1 1 1 1
pfem 0 0 0 1 1 1
pmal1 9 9 9 9 9 9
pmal1 0 0 0 1 1 1
pmal2 9 9 9 9 9 9
pmal2 1 1 1 0 0 0

[AprilYUZhang]

The result of hap_0.5 value

f 9 9 9 9 9 9
f 1 1 1 1 1 1
ff 9 9 9 9 9 9
ff 0 0 0 0 0 0
fm 1 1 1 1 1 1
fm 0 0 0 0 0 0
m 1 1 1 1 1 1
m 0 0 0 0 0 0
mf 9 9 9 1 9 9
mf 1 1 1 1 1 1
mm 0 0 0 0 0 0
mm 0 0 0 0 0 0
pfem 1 1 1 1 1 1
pfem 0 0 0 1 1 1
pmal1 9 9 9 9 9 9
pmal1 0 0 0 1 1 1
pmal2 9 9 9 9 9 9
pmal2 1 1 1 0 0 0

and segregation file

f 0.6080 0.6800 0.8000 0.9999 0.8000 0.6800
f 0.3920 0.3200 0.2000 0.0001 0.2000 0.3200
f 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
f 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
ff 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
ff 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
ff 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
ff 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
fm 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
fm 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
fm 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
fm 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
m 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
m 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
m 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000
m 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000
mf 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
mf 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
mf 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
mf 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
mm 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
mm 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
mm 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
mm 0.2500 0.2500 0.2500 0.2500 0.2500 0.2500
pfem 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
pfem 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
pfem 0.5942 0.6569 0.7616 0.9359 1.0000 1.0000
pfem 0.4058 0.3431 0.2384 0.0641 0.0000 0.0000
pmal1 0.0000 0.0000 0.0001 0.7853 1.0000 1.0000
pmal1 1.0000 1.0000 0.9999 0.2147 0.0000 0.0000
pmal1 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
pmal1 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
pmal2 1.0000 1.0000 0.9999 0.2147 0.0000 0.0000
pmal2 0.0000 0.0000 0.0001 0.7853 1.0000 1.0000
pmal2 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
pmal2 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

[AprilYUZhang]

Firstly, I highlight the segregation and hap_0.5 of "pfem" , the exact cross point is between third and forth, which estimates correctly in hap_0.5. But segregation prob looks ambiguous. This diff comes from alphapeel cannot identify the cross happened on "m"
Second problem: If three offspring have recombination, this recombination is more likely to happen on their mother. But the results show AlphaPeel doesn't have this preference. I have to emphasize the importance of preference, because it must affect segregation prob that mislead to estimate recombination rate and cross point.
I think it might be a reason why Yapp outperforms AlphaPeel in recom estimate, because Yapp infers recomb point based on offspring's genotype in one family.

m 1 1 1 1 1 1 # expext 0 0 0 1 1 1
m 0 0 0 0 0 0 # expect 1 1 1 0 0 0
pfem 1 1 1 1 1 1 # expect 1 1 1 1 1 1
pfem 0 0 0 1 1 1 # expect 0 0 0 1 1 1

m 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
m 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
m 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000
m 0.5000 0.5000 0.5000 0.5000 0.5000 0.5000
pfem 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
pfem 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
pfem 0.5942 0.6569 0.7616 0.9359 1.0000 1.0000
pfem 0.4058 0.3431 0.2384 0.0641 0.0000 0.0000

[gregorgorjanc]

Nice example @AprilYUZhang. I agree with your summary, the recombination should have been inferred in the mother, not three times in her children. However, is there information for this to happen in the mother already?

But, is this actually a legit example? Let me start with genotypes for mother and her parents. You had

mm 0 0 0 0 0 0
mf 1 1 1 1 1 1
m 1 1 1 1 1 1

and you assumed

m 0 0 0 1 1 1
m 1 1 1 0 0 0
mf 9 9 9 9 9 9
mf 1 1 1 1 1 1
mm 0 0 0 0 0 0
mm 0 0 0 0 0 0

BUT, mm could only have passed onwards haplotype 0-0-0-... to her daughter, while mf could have only passed on haplotype 1-1-1-... to his daughter - recall this is the X chromosome.

So, with this constraint in mind (and this is what your modified X chromosome Mendelian segregation function does!), the only possible explanation for the m 1 1 1 1 1 1 genotype is that we got 00000 from mm and 111 from mf, which is what AlphaPeel gave you;) So, it's actually doing the right thing, and then it of course pushes the recombination to progeny;)

So, you will have to change the example a bit more to test if recombination works well.

I noticed 999991999 haplotype in the output for a male. It would be good to know where that 1 is coming from - I thought that we fixed that last time?

[AprilYUZhang]

In theory, I agree with you. I give an ambiguous example on purpose to test how it works with recombination. I believe alphapeel will work well in a simple and clear example. I just want to highlight what problem we face and find a path to improve recombination estimates, which is important for TS inference. This means our peel-down is good, but it will be better if peeling up can give more information under this situation.
Alphapeel cannot achieve this preference in principle because it does have these two possibilities, one occurs in “m” and the other occurs in her offspring. In this example, which of these two cases is more likely? When we observe this data, the probability that the first of these two cases will occur is greater than the second, which will be a very large performance improvement if alphapeel can give some clues in the segregation prob. And we also did peel up. In principle, the situation of the offspring should also affect the seg probability of the previous generation. However, in the current result file, the seg probability of the previous generation was almost unaffected.
Another thing is whether this fuzzy inference has a great impact on the inference of recombination？ The answer is very large. If alphapeel does not have this preference, the cross actually only happens once, but we will infer 3 recombinations from segregation prob.

[AprilYUZhang]

In terms of "999991999", I used the old version, here is the new result.

f 9 9 9 9 9 9
f 1 1 1 1 1 1
ff 9 9 9 9 9 9
ff 0 0 0 0 0 0
fm 1 1 1 1 1 1
fm 0 0 0 0 0 0
m 1 1 1 1 1 1
m 0 0 0 0 0 0
mf 9 9 9 0 9 9
mf 1 1 1 1 1 1
mm 0 0 0 0 0 0
mm 0 0 0 0 0 0
pfem 1 1 1 1 1 1
pfem 0 0 0 1 1 1
pmal1 9 9 9 9 9 9
pmal1 0 0 0 1 1 1
pmal2 9 9 9 9 9 9
pmal2 1 1 1 0 0 0

[gregorgorjanc]

In theory, I agree with you. I give an ambiguous example on purpose to test how it works with recombination. I believe alphapeel will work well in a simple and clear example. I just want to highlight what problem we face and find a path to improve recombination estimates, which is important for TS inference. This means our peel-down is good, but it will be better if peeling up can give more information under this situation. Alphapeel cannot achieve this preference in principle because it does have these two possibilities, one occurs in “m” and the other occurs in her offspring. In this example, which of these two cases is more likely? When we observe this data, the probability that the first of these two cases will occur is greater than the second, which will be a very large performance improvement if alphapeel can give some clues in the segregation prob. And we also did peel up. In principle, the situation of the offspring should also affect the seg probability of the previous generation. However, in the current result file, the seg probability of the previous generation was almost unaffected. Another thing is whether this fuzzy inference has a great impact on the inference of recombination？ The answer is very large. If alphapeel does not have this preference, the cross actually only happens once, but we will infer 3 recombinations from segregation prob.

I think I know what you mean and we could look into way to “force” recombination higher, however the issue is that your example genotypes do not match with what you assumed has happened with haplotypes, since this is an X chromosome. To see this, look only at

mm 0 0 0 0 0 0
mf 1 1 1 1 1 1
m 1 1 1 1 1 1

How can m (mother/female) inherit an allele 1 from mm (her mother) if this is an X chromosome? It cannot, so AlphaPeel having only this information, has to conclude that she inherited 0000000 haplotype from mother (that's the only possible haplotype SHE can pass to his children). and 111111 haplotype for father (that's the only possible haplotype HE can pass to his children).

Do you see this?

As a next stage I suggest you create an example with recombination that is biologically possible first, then we think about more advanced recombination inference/fixing techniques;)

[XingerTang]

@AprilYUZhang @gregorgorjanc Do you have any updates on how this conversation goes

[gregorgorjanc]

The issue here is that @AprilYUZhang created biologically impossible example. As suggested in #190 (comment) @AprilYUZhang create biologically realistic example as a test. If you struggle with this, let's do it together.

[AprilYUZhang]

Yes, this has been solved, close now.

[gregorgorjanc]

Let's get into habit into linking where (in which commit/PR) has something been solved. At the moment I cannot check this easily. Can you provide a link @AprilYUZhang ?