Announcement

Collapse
No announcement yet.

Accuracy of Gedmatch Utilities

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Accuracy of Gedmatch Utilities

    Just curious about what everyone's experience on Gedmatch has been with the Eye Colour and 'Are Your Parents Related' tools. I uploaded my grandmother's data today. The eye colour was completely off - it gave her light blue, while her eyes are actually dark brown. As for the 'Parents' tool, it gave her zeroes across the board - but her parents were, in fact, second cousins once removed!

  • #2
    The eye color was mostly right on two of the three tests I am administering.

    Comment


    • #3
      Well... the eye color got my mother at least approximately right (brown eyes), but it has been awhile since I looked into her eyes closely, so all I can say on that is that I think hers might be a little darker. They do say that FTDNA is less accurate on that utility due to not including medically significant areas.

      I wasn't going to do anything with the other one as my mother's parents would not be related in any near period, but to my surprise it indicated that there could be some (probably fairly old) connection. I'd guess that it dated back to Vikings preying on Scotland, but you never know...

      What has been interesting is all the admixture utilities. Looks like my mystery great-grandfather may have been part Native American. Either that or the Norwegians had some Asian ancestry. But I tried most of the admixture utilities and they generally either listed Native American or surprising amounts of different Asian areas.

      It's pretty neat to be able to do all this at last.

      Comment


      • #4
        Eye color Prediction:
        CC at: rs3794604 - Blocks some melanin. Often gives light colored eyes.
        GG at: rs7174027 - Blocks some melanin. Often gives light colored eyes.
        CC at: rs4778241 - Low Melanin. Basis for Gray, Blue, Green, or Yellow Eyes if no other pigmentation is present.
        CC at: rs9782955 - Blocks some melanin. Often gives light colored eyes.
        TC at: rs1129038 - Adds Yellow.
        GG at: rs12906280 - Gray ring around outer edge
        CC at: rs7403602 - High density on Anterior Stroma. Blocks melanin. Blocks blue. Gives lighter colors.
        AA at: rs11634406 - Flecks (Nevi).
        I have dark brown eyes.


        Parents related Results:
        Largest individual segment of SNPs with greater than 500 matches: 0
        Total autosomal SNPs in segments with greater than 500 matches: 0
        Largest Single-Block Genetic Distance = 0 cM
        Total Genetic Distance = 0 cM
        Generally, when the Largest Single-Block Genetic Distance is less than 1, it can be assumed that no relationship exists.
        They're 8th-11th cousins 15 ways and about 5% of my family finder matches appear on both sides of the family.

        Comment


        • #5
          Originally posted by AlisonT View Post
          Just curious about what everyone's experience on Gedmatch has been with the Eye Colour and 'Are Your Parents Related' tools. I uploaded my grandmother's data today. The eye colour was completely off - it gave her light blue, while her eyes are actually dark brown. As for the 'Parents' tool, it gave her zeroes across the board - but her parents were, in fact, second cousins once removed!
          The last time I ran the eye color predictor on my data, it was dead-on for 23andMe data. For FT-DNA Family Finder, it got the basic structure right but it missed the color.

          Comment


          • #6
            I have never gotten the eye color one to work correctly, but i have been using only FTDNA files. I will try it again with Ancestry and 23andme and see if the prediction improves.

            The "Cousin Test For Parents" function seems to work for me, here is what it said on my kit:

            Largest individual segment of SNPs with greater than 500 matches: 904
            Total autosomal SNPs in segments with greater than 500 matches: 1709
            Largest Single-Block Genetic Distance = 1.577 cM
            Total Genetic Distance = 1.6692 cM

            I know my parents are something like 14th or 17th cousins - probably several times over due to endogamy (Colonial VA). I also tried it on my brother-in-law's kit (Ashkenazi Jew), it looks like it's confused by the endogamy because I've done his genealogy and I'm reasonably sure his parents are not 6th cousins. They might well, however, be 9th cousins twice over, or the like - I haven't gotten that far back yet on all the lines to rule it out.

            Largest individual segment of SNPs with greater than 500 matches: 1658
            Total autosomal SNPs in segments with greater than 500 matches: 7852
            Largest Single-Block Genetic Distance = 7.641 cM
            Total Genetic Distance = 21.0863 cM

            What I want to know about accuracy are the Admix utilities, for example if I run my results (largely Colonial American with some Swiss Mennonite and a recent possible NPE) through the Dodecad World9 Calculator, by Chromosome, and it tells me I have between 0.3 and 3.0 percent Amerindian, scattered across my chromosomes, is that even remotely reliable? My half-brother has some too, though less of it and on fewer chromosomes. I would say I have approximately double the percentages he has. I do have the typical "Indian Princess" legends on my father's side of the family, and my mother's mother was adopted, I know her biological mother but the father is most likely NOT the man whose name she put on the birth certificate. So it could be coming from either side. Any thoughts?

            Comment


            • #7
              Originally posted by brookes View Post
              I have never gotten the eye color one to work correctly, but i have been using only FTDNA files. I will try it again with Ancestry and 23andme and see if the prediction improves.
              It appears that we don't pay attention to what is right in front of our eyes. The follows text appears on the GEDmatch.com Eye Color Prediction Kit No. entry webpage:
              23andMe V3 results work best. V2 will work, but may miss the prediction more often.

              FTDNA Illumina results may provide a prediction, but generally there are not enough of the necessary SNPs for it to be accurate.

              FTDNA Affymetrix results do not work - not enough SNPs.
              My reading [at long last] of this text is that we have no reasonable expectation that the Eye Color Prediction tool will work with FT-DNA Family Finder data. I got around to reading this text last night after I uploaded my AncestryDNA data to GEDmatch.com. The prediction using my AncestryDNA looks just like the prediction using my Family Finder data--wrong. I infer that AncestryDNA like FT-DNA Family Finder does not measure enough SNPs for an accurate prediction.

              ...

              What I want to know about accuracy are the Admix utilities, for example if I run my results (largely Colonial American with some Swiss Mennonite and a recent possible NPE) through the Dodecad World9 Calculator, by Chromosome, and it tells me I have between 0.3 and 3.0 percent Amerindian, scattered across my chromosomes, is that even remotely reliable? My half-brother has some too, though less of it and on fewer chromosomes. I would say I have approximately double the percentages he has. I do have the typical "Indian Princess" legends on my father's side of the family, and my mother's mother was adopted, I know her biological mother but the father is most likely NOT the man whose name she put on the birth certificate. So it could be coming from either side. Any thoughts?
              Over the last few weeks, I have been on a mission concerning this issue:
              1. Each admix model will give a result for your genome.
              2. You should seek the most accurate results. However, you cannot predict the accuracy of a model for your specific genome based on the name of the model or your current understanding of your genetic heritage.
              3. The accuracy of the model can only be determined after you run your data through it. To determine its accuracy, run the oracle for the model and take note of its 4-Populations Genetic Distance. The smaller the Genetic Distance, the more accurate the results.
              4. In the case of a model that does not have an oracle, its accuracy cannot be determined.

              Run your data through each model available on GEDmatch.com. Accept the results of the model that yields the smallest Genetic Distance for your data.

              Comment


              • #8
                Yes, I know that FF results do not work as well with the eye colour predictor. I did read that. I wasn't criticizing Gedmatch, just curious about what other's results had been.

                I still think it's strange that my grandmother gets all zeroes on 'Are Your Parents Related'. You'd think if your parents were second or third cousins, it would be bound to show up.

                Comment


                • #9
                  Originally posted by AlisonT View Post
                  I still think it's strange that my grandmother gets all zeroes on 'Are Your Parents Related'. You'd think if your parents were second or third cousins, it would be bound to show up.
                  It sounds like you may need a second opinion. David Pike's ROH utility is very good. He also offers links to some research papers to help you interpret the results.

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

                  Comment


                  • #10
                    I posted earlier that the gedmatch eye color predictor had been correct in 2 of three of the tests I had submitted. Those were FTDNA tests. I also handle several Ancestry.com tests which I uploaded to gedmatch this morning. At least in my case, the Ancestry.com tests seem to be working quite well with the eye color predictor. Four of 4 have been correct. In one of those cases the test was a duplicate of an FTDNA test (i.e. the person had tested with both FTDNA and Ancestry). For whatever reason, the FTDNA test results caused the eye predictor to estimate blue eyes when the person had brown eyes. The Ancestry test of this same person, when ran though the predictor, came out with the correct brown eye estimate. This is of course only one incident, but it may be the Ancestry samples are a better fit to the gedmatch eye predictor utility.
                    In any event gedmatch is a great service!

                    Comment


                    • #11
                      Originally posted by Swennilsson View Post
                      In any event gedmatch is a great service!
                      Indeed.

                      It is interesting that they are able to outperform a commercial company such as FTDNA considering these differences:

                      1) Unpaid staff,
                      2) Open-source software, and
                      3) Reverse engineering of the format of the data they process.

                      (I speculate that they are currently keeping themselves busy with no. 3).

                      Comment


                      • #12
                        Gedmatch got my eye color very accurately using FTDNA results.

                        Early on, I found the "are your parents related" function very useful for determining that I have no (or possibly <1%) Ashkenazi ancestry on my paternal side (there was a tiny bit of shared DNA). This conflicted with what Dr. McDonald estimated (>55% Jewish) but is consistent with everything else I've been able to determine and with 23andMe's Ancestry Finder results. By contrast, my half-brother's results for parentally relatedness were high, and his partially phased results (using my DNA) were quite consistent with the calculator and consistent with his being fully or virtually fully Ashkenazi.

                        Comment


                        • #13
                          Hi JOLson

                          I have just run my sons results through the link you provided. I wondered if you could help me understand the 3 statements below

                          Total autosomal (Chr 1-22): 0.498 % ( 3528 of 708092 SNPs) are NoCalls
                          Total autosomal (Chr 1-22): 29.219 % (206900 of 708092 SNPs) are Heterozygous (this tally excludes 0 heterozygous SNPs that were treated as homozygous)
                          Total autosomal (Chr 1-22): 70.781 % (501192 of 708092 SNPs) are Homozygous (this tally includes 0 heterozygous SNPs that were treated as homozygous)
                          Processing Completed.

                          Comment


                          • #14
                            I'm not an expert on David's ROH utility, but I'll take a stab at it...

                            No-Calls are positions where FTDNA was unable to determine the actual markers, for one reason or another. They are generally ignored.

                            Heterozygous means the 2 DNA "letters" are different.

                            Homozygous means the 2 DNA letters are the same.

                            The statement that "x Heterozygous SNPs were treated as Homozygous" is an allowance for the occasional error in the data that inevitably slips in. In this case it ignores heterozygous SNPs that are at least 150 SNPs apart.

                            Any Runs of Homozygosity (ROHs) of a significant size would be an indication that this person's parents were related.

                            Comment

                            Working...
                            X