cross-posted from: https://kbin.social/m/news@lemmy.world/t/895673

DNA My Dog received human genetic sample and identified it as a malamute, shar-pei and labrador, according to news station

  • Sibbo@sopuli.xyz
    link
    fedilink
    arrow-up
    27
    ·
    8 months ago

    This may be easy to explain in case their system doesn’t account for the possibility that whatever DNA they get is not from a dog. Then whatever they answered is just the dog race with DNA closest to human, according to their metrics.

    They very likely have a set of possible target DNA references, and look at how close whatever they get is to each of these.

    And there are different metrics with which one can measure DNA similarity. Hence it also is to be expected that a different company would return different results. Albeit not super different.

    • Ajen@sh.itjust.works
      link
      fedilink
      arrow-up
      12
      ·
      8 months ago

      They also don’t report the confidence of their results. DNA tests, even the kind used for forensics in criminal investigations, don’t give perfect results, there’s always some uncertainty. I think there are probably a lot of instances where they have a very low confidence in the results but send them out anyway.

      • Sibbo@sopuli.xyz
        link
        fedilink
        arrow-up
        8
        arrow-down
        1
        ·
        8 months ago

        As of today, it is impossible to report an accurate confidence value for these kinds of analyses. The required theoretical work has not been done yet, and of course may turn out impossible. However, there are tons of ways to estimate confidence.

        One of the other companies cited in the article seems to be a bit better about this. They seem to filter out non-dog DNA, and if there is not enough dog DNA left after this, then they report that the sample was bad.

    • JoBo@feddit.uk
      link
      fedilink
      arrow-up
      8
      arrow-down
      3
      ·
      8 months ago

      Dogs have 38 pairs of chromosomes, humans only 23. It should not be possible for this to happen.

      In the article they claim they got results from one of the human cheek swabs and not the other. It is possible that they just don’t keep their equipment clean enough and some actual dog DNA got into the mix.

      Either way, the other companies receiving similar samples just said there was no dog DNA there. This company has repeatedly provided dog results for human DNA samples. Whatever happened they’re doing something very wrong.

        • JoBo@feddit.uk
          link
          fedilink
          arrow-up
          6
          arrow-down
          1
          ·
          8 months ago

          If I’m wrong, I don’t know how I went wrong so please, do me a favour and explain.

            • JoBo@feddit.uk
              link
              fedilink
              arrow-up
              2
              arrow-down
              1
              ·
              edit-2
              8 months ago

              I’m not sure what an article about whole genome sequencing has to do with anything? These companies are looking at a handful of SNPS.

              Can you explain how it is possible to look at an SNP without knowing which chromosome it is located on?

              And how the other companies were confident there was no dog DNA in the sample and this company said it found dog DNA in only one of the two samples sent?

              • Sibbo@sopuli.xyz
                link
                fedilink
                arrow-up
                3
                arrow-down
                1
                ·
                8 months ago

                The article is about sequencing technology. This is what is needed to get from physical DNA to its textual representation. The assignment to chromosomes is lost in the process. It can either be reconstructed after sequencing, by e.g. whole genome assembly or k-mer based methods.

                But there are other methods of identifying races than looking at correctly identified SNPs. Often, k-mer content is enough. And it is likely more efficient, because it does not require any computationally heavy assembly.