Affy to Illumina conversions complete?

[swimmingly]

conversions not 100% completed

Turns out conversions are not all done, and my missing match hasn't yet been converted, so I am actually NOT missing my third cousin and look forward to seeing him on the Illumina eventually. So that is all good, and FTDNA was very good about responding to my query.

[JPHutchins]

I got two new matches today, neither of them were conversions from Affy.

One, however, was indicated to be a probable 4th cousin, the other was 5th to distant.

[bkilpatrick]

I got two new matches today, neither of them were conversions from Affy.

One, however, was indicated to be a probable 4th cousin, the other was 5th to distant.

Same here.

[Red Rover]

My new match responded to me and he is an new Illumina not an Affy conversion.

As an aside I am going to curse the 19th century English government for the poor state of Irish records going back past the mid 19th century. Another match from Ireland who I probably won't be able paper match.

My commoner English ancestors are cool though.

[189786]

If you call, or contact by email, you should be prepared to send your Affy match file, or both match files. I combined mine and added column for source (Affy or Illumina) so it could be easily sorted to see the missing matches. Are you saying that I should put both lists of matches into a spread sheet? If not, please explain--I don't understand.

If you have a couple of examples and can provide kit numbers of clearly related lost matches that might also help.

How do I know what someone's kit number is unless I ask the person? I could tell FTDNA the names of my two matches but I don't think I know their kit number.

[nolnacsj]

Yes I have lost three Affy matches with over 10cM
3rd Cousin - 5th Cousin 36.57 shared, 12.71 longest block
4th Cousin - Distant Cousin 34.29 shared, 11.67 longest block
4th Cousin - Distant Cousin 33.76 shared, 10.04 longest block

19/28 of my illumina matches are below 10.0 cM longest block

I have a few Affy's with a longest block over 10 cM which have not converted also.

I did finally get one new Illumina match today, a weak 5th-distant cousin match. First new match since 4/19. Affy's still missing.
Judy

[gtc]

How do I know what someone's kit number is unless I ask the person? I could tell FTDNA the names of my two matches but I don't think I know their kit number.

If you administer a project then you can see the kit numbers of the members of that project. Apart from that you need to ask the individuals.

I don't know if FTDNA will accept names without kit numbers in your particular circumstance, but you can always ask them.

[Ann Turner]

Turns out conversions are not all done, and my missing match hasn't yet been converted, so I am actually NOT missing my third cousin and look forward to seeing him on the Illumina eventually. So that is all good, and FTDNA was very good about responding to my query.

The case I mentioned (with two missing close cousins) had results come in last night.

Many of the cases involving more distant cousins turn out to be what I'd call false positives when I look at the raw data. The Affy chip had a higher error rate, and FTDNA compensated for this by tolerating a few mismatches here and there in the middle of a long consecutive run of matching SNPs. The Illumina chip has a lower error rate plus more SNPs, so it ferrets out more of the false positives.

I do have a quirky example of a true positive Affy match that didn't pass the Illumina test. I'll post more on that when I've finished analyzing it.

[mkdexter]

The Illumina chip has a lower error rate plus more SNPs, so it ferrets out more of the false positives.

I asked FTDNA yesterday if the error tolerance setting in Illumina's algorithm is lower or the same as Affy's setting. Still waiting for an answer on that question.

[JohnG]

It is sort of like medical tests, how many false positives do you tolerate in order not to ignore some real positives?

In matches a 'false positive' probably means it is still a match but just more distant.

I attached a graph from some known relationships data posted on this forum from Gedmatch. I put my data on it too. My 1st, 1st 1 R, and 2nd cousins show as circles at top right. The 3rd to 5th known relations spread out a lot.

If they make matches more certain, some matches may be lost that are closer than some that are kept. But there will be fewer overstated matches, I guess.

[Ann Turner]

It is sort of like medical tests, how many false positives do you tolerate in order not to ignore some real positives?

In matches a 'false positive' probably means it is still a match but just more distant.

I attached a graph from some known relationships data posted on this forum from Gedmatch. I put my data on it too. My 1st, 1st 1 R, and 2nd cousins show as circles at top right. The 3rd to 5th known relations spread out a lot.

If they make matches more certain, some matches may be lost that are closer than some that are kept. But there will be fewer overstated matches, I guess.

Yes, it's always a trade-off between false positives and real positives.

However, some of the false positives we see using *genotype* data are actually pseudo-segments, not even reflecting more distant ancestry. They are patched together with contributions from the paternal side alternating with contributions from the maternal side. A match that looks like a single segment can fracture into dozens of very short pieces if you look at *haplotypes*, where you've determined which parent contributed which allele. This is practical only if you have raw data for father/mother/child trios, the reason I was asking for datasets of that type.

[aidan]

One of my 'Close and Immediate' cousins on the Affy chip are not close cousins on the Illumina chip.

Affy -
Relationship Range: 3rd Cousin - 5th Cousin
Suggested Relationship: 4th Cousin
Shared cM: 37.46
Longest Block: 13.42

Illumina -
Relationship Range: 4th Cousin - Distant Cousin
Suggested Relationship: -
Shared cM:28.84
Longest Block: 12.78

[JohnG]

Yes, it's always a trade-off between false positives and real positives.

However, some of the false positives we see using *genotype* data are actually pseudo-segments, not even reflecting more distant ancestry. They are patched together with contributions from the paternal side alternating with contributions from the maternal side. A match that looks like a single segment can fracture into dozens of very short pieces if you look at *haplotypes*, where you've determined which parent contributed which allele. This is practical only if you have raw data for father/mother/child trios, the reason I was asking for datasets of that type.

One of my major learnings from Family Finder is how interrelated all of us are from even a vague huge grouping like Americans whose families have been here 250 years. I find a 5th cousin and they are also my wife's 4th cousin. My wife is my 1st cousins 5th cousin, etc. And I am talking of paper trails as well as dna matches.

The genes come together and split and come together later on. The pseudo-segments have to be the same in several people to make a match. That means the accidental reshuffling has happened in at least two cases. I think I may have cases where it showed up in a multiple cases - some of those 8 cM segments that a dozen matches shared. So they are pseudo in that they do not come from one individual to their child, but maybe they come from a shared ancestral population down to many descendents?

[Ann Turner]

quirky case

I've mentioned that a goodly number of Affy matches that don't show up in Illumina could very well be false positives.

On the flip side, I have an example I would consider a true positive, but it didn't pass the Illumina test due to a "perfect storm" of circumstances. Much of this analysis was expedited by David Pike's utilities at

http://www.math.mun.ca/~dapike/FF23utils/het.php

It was a robust 12 cM, 8.75 Mb match with about 2500 SNPs, originally found at 23andMe and confirmed in the Affy test at FTDNA. I had also checked it with some haplotype data, not just genotypes.

The 23andMe test had no discrepancies in the region, while the FTDNA had five. This illustrates why FTDNA allowed a certain number of mismatches to compensate for the error rate on the Affy chip (still low in absolute terms).

The FTDNA Illumina chip added more SNPs, and the original segment split in two parts, 6.11 Mb (no mismatches at all) and 2.75 Mb. These two pieces were separated by just 9500 bases and had three mismatching SNPs in a row. That can be a sign of a microdeletion, where one of the parties actually has just one allele, but it is reported as a homozygous genotype.

The original test had just one SNP in the "microdeletion" region, and it was a match. However, the allele is very common, found in 95% of Europeans, so it's not surprising that it passed muster.

The 6.11 Mb section in the new Illumina test is 7.65 cM in length, so presumably it just missed the FTDNA threshold, which seems to be 7.7 cM. (The cM value is according to the Rutger's map, which may or may not be what FTDNA is using but is probably close).

So bottom line: the additional Illumina SNPs revealed a microdeletion, which chopped the segment into two parts, and the larger part just missed a threshold.

I'm posting this, not to get your hopes up about all of your missing Affy matches, but to illustrate that it sometimes takes digging down deep in the raw data to understand what's going on. There's no way to develop an algorithm to handle quirky cases like this.

[Ann Turner]

One of my major learnings from Family Finder is how interrelated all of us are from even a vague huge grouping like Americans whose families have been here 250 years. I find a 5th cousin and they are also my wife's 4th cousin. My wife is my 1st cousins 5th cousin, etc. And I am talking of paper trails as well as dna matches.

The genes come together and split and come together later on. The pseudo-segments have to be the same in several people to make a match. That means the accidental reshuffling has happened in at least two cases. I think I may have cases where it showed up in a multiple cases - some of those 8 cM segments that a dozen matches shared. So they are pseudo in that they do not come from one individual to their child, but maybe they come from a shared ancestral population down to many descendents?

What you say is conceptually possible. We need a better vocabulary, but perhaps we could call those overlapping segments in distinction to pseudo-segments, where I am talking about tiny bits and pieces. But we really need phased data to demonstrate what's happening.