Archiver > GENEALOGY-DNA > 2010-02 > 1265959447

From: "Lancaster-Boon" <>
Subject: [DNA] Variance Assessment of R:U106 DYS425Null Cluster
Date: Fri, 12 Feb 2010 08:24:07 +0100

Dear Anatole,

I'll put responses which are about my answers instead of the subject on to a
mail to you directly. I note that our moderator is being stricter lately.

>Andrew:...So, I think I can summarize as follows:

>1. SOMETIMES, in second order or complicated cases, you need to separate
out sub-branches and handle them seperately.

Anatole: I do not know why you put an emphasis on SOMETIMES. Of course it is
sometimes, how else? Then, there is no "second order" here. "Second order"
has a completely different meaning, being compared with "first order". It is
not applicable for these mutations. The rest is correct.


>Andrew: 2. To do this, you use, I am now reasonably confident, the various
normal techniques for this, either by eye or using one of the software
solutions, or may be your own? I don't hear of any new technique here?

Anatole: I repeat, that I would not recommend to do it "by eye". [snip]

I think this means you basically agree with my description above?

Andrew: If I understand correctly, it is easy to define an extreme case
"first order" example of data as per your description. It is a comb shaped
tree, or star shaped network...

Anatole: So far, so good... But you forgot again the principal criterion -
the logarithmic method. "A comb shaped tree" is hard to find. Some
not-so-comb-shaped trees can still have one common ancestor. There are VERY
MANY transient cases.


>Andrew: So (my new question NB) how do you decide when the branching is
important for the analysis and when not?

Anatole: I have explained above, probably fifth or sixth time. I apply the
dual criterion. I compose a tree. There are branchings and branchings. Some
of them are distinct and clearly have a separate base haplotype each. Some
of them look like a fish skin, they are part of the same branch.

Yes, you make a tree, and everything depends on THAT step, as you also said.
But you have much new to say about THIS step, except that it happens. So all
such trees are, as I said, a shot in the dark, the same as they are for
everyone else. So now I think I see more clearly that what you are saying is
that this step for you is not part of your method but just assumed to work.
For you it is apparenlty a black box which you plug into your method. For
other people it is the most critical step which can go wrong. You are
looking at what constitutes the calculation from a different perspective to
for example Ken.

Note by the way that the biggest effect in terms of concrete result is not,
as far as I can see, any difference in your estimation of the most likely
answer, but in your estimation of how big the confidence interval is for
intraclade calculations.

I think you missed this in what Ken was saying to you.

>Andrew: Personally I thought my example was a good one.

Anatole: No. Though, in a way, it was a good one to show that it was a wrong

It was a real one though. I take it this effectively means you are saying it
just a very unlucky and difficult case that I have found. As a person
working with a lot of family groups I do not agree though. It was just one I
was working on that day.

>Andrew: I guess one possible tree they'll come up with with group the set
into 4 possible combinations (18 and 11, 18 and 12, 19 and 11, 19 and 12)...

Anatole: Do not guess. This is how people get confused.

Are you saying no computer programme can come up with such answers?

I think you are missing my point which is that you have to guess or else
find a mechanical method of guessing. In this part of the analysis, making
your "haplotype tree", using a neutral "method" or a software programme does
not stop it still being a shot in the dark or guess like the rest of make.
It maybe just feels like it is not, just like tossing a coin is a way of
saying that no individual made the decision, but it is still a guess, and it
is still the most critical part of your method, and it still playing a role
in how confident we should be in the evnetual answer.

Best Regards

This thread: