GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2009-12 > 1260131591


From: "Tim Janzen" <>
Subject: [DNA] FW: real relatedness or fluff and fairytales?
Date: Sun, 6 Dec 2009 12:33:11 -0800


Dear All,
In general, I agree with Thomas's words of caution about use of the
Relative Finder feature. However, my overall impression of the 23andMe
Relative Finder is this: "It works!". If you are closely related to
someone at least at the 2nd or 3rd cousin level of relationship, then this
will be obvious from the Relative Finder feature or use of the Family
Inheritance feature. If there is a somewhat more distant genealogical
relationship, then it will also show up in the results.
For the past week or so I have been thinking about the possibility
that errors in the SNP calls on 23andMe's chips could lead to a breaking of
the chain of SNPs that are IBD. This is one of the reasons that I have been
collecting data on the differences between the total segments and cMs given
using Wayne Kauffman's modified version of Jim McMillan's program on the raw
23andMe SNP data and the total segments and cMs given in 23andMe's Advanced
Family Inheritance function found under the Ancestry Labs tab. I wanted to
see if there were significant differences in the cM totals between the two
different methods. Below are 6 columns of data. The columns are results
from comparing my dad's raw 23andMe data to the raw data from 14 other
people of Low German Mennonite ancestry who are not more closely related to
him than the 4th cousin level of relationship. The results from Wayne's
analysis can also be seen at
http://kquilting.homeserver.com/23andme/results.html. The columns headings
are as follows:

1. total cMs for segments longer than 4 cM as per Wayne Kauffman's version
of Jim McMillan's program

2. total cMs for segments longer than 5 cM as per Wayne Kauffman's version
of Jim McMillan's program

3. total cMs of Half Identical by Descent segments in 23andMe's Advanced
Family Inheritance function (These segments are longer than 5 cM)

4. total number of segments greater than 4 cM in length as per Wayne
Kauffman's version of Jim McMillan's program

5. total number of segments greater than 5 cM in length as per Wayne
Kauffman's version of Jim McMillan's program

6. total number of HIR (blue) segments appearing in a direct comparison in
23andMe's Family Inheritance feature

99.0 85.1 82 10 7 7
115.0 97.2 91 12 8 8
59.0 59.0 53 5 5 4
76.4 53.7 56 11 6 7
61.9 44.0 41 7 3 3
60.6 43.5 41 7 3 3
80.9 71.7 45 12 10 6
47.1 47.1 39 5 5 4
67.4 55.0 55 7 4 4
49.6 33.1 31 6 2 2
35.4 31.3 29 4 3 3
53.3 35.9 23 9 5 3
37.5 28.4 22 5 3 2
83.0 51.1 34 13 6 3

The important thing to note here is that totals in columns 2 and 3
are relatively consistent with each other with the exception of one
individual. This means that whatever method that 23andMe is using to
determine the lengths of the HIRs is producing similar outcomes to whatever
method Wayne Kauffman's version of Jim McMillan's program is using. I have
noted that in general Wayne Kauffman's version of Jim McMillan's program
generally gives a longer length in cMs for HIRs than does 23andMe's Advanced
Family Inheritance function. This means that in general 23andMe's technique
is more conservative than Jim McMillan's program's program is in both
calling HIRs and in determining their length. I haven't studied Jim
McMillan's program in-depth up to this point and I thus can't comment as to
whether or not Jim McMillan's program takes into account individual SNP
errors as it is determining the lengths of the HIRs. I am copying Wayne and
Jim on this message and perhaps they will have additional insights.
One thing I don't like about 23andMe's Advanced Family Inheritance
function is that 23andMe rounds the starting and ending points on the
chromosome for the HIRs to the nearest millionth base pair position. I wish
that they would report the actual base pair position like Jim McMillan's
program does. I also agree with Thomas that 23andMe should provide
information about any inconsistencies raw SNP data for the HIRs that they
are reporting.
Sincerely,
Tim Janzen


-----Original Message-----
From:
[mailto:] On Behalf Of Thomas Krahn
Sent: Sunday, December 06, 2009 11:17 AM
To:
Subject: Re: [DNA] real relatedness or fluff and fairytales?

Also there is apparently a threshold for "wrong SNP" calls. If you
compare a parent-child result closely you will find a couple of SNPs
that don't share any of the alleles. 23andMe simply considers them as
typing errors or mutations and ignores them. This is in my opinion very
dangerous (especially in highly conserved regions). The correct way to
handle this would be to at least list the inconsistencies and if
possible give a reason why they are ignored. So for now I recommend to
not simply trust the display of a continuous haploblock in the 23andMe
chromosome browser. You should definitely verify the raw SNP calls.

So generally speaking the the relative finder is an extremely simplified
tool to detect a possible related person. An experienced DNA genealogist
will certainly want to have a deeper look into the real raw results
before he blindly accepts the relative finder output.

Thomas

This thread: