Page 2 of 5 FirstFirst 1234 ... LastLast
Results 11 to 20 of 46

Thread: Cheddar Man near totally debunked (the "blue eyed black skinned" ancient Briton)

  1. #11
    Registered Users
    Posts
    74
    Sex
    Y-DNA (P)
    G-M406
    mtDNA (M)
    M7c1c3

    Quote Originally Posted by Matp View Post
    So the effect of rs1800414 is actually not huge- compare what happens to Loschbour and Cheddar man predictions when you add a single allele to each prediction (outlined in red):

    Attachment 47065Attachment 47066

    So the addition of a single allele on each prediction (responsible for a 1.3 melanin unit reduction) results in Loschbour getting 0.10 lighter in the intermediate category- whilst Cheddar man goes from 0.85 dark to black to 1.0 for intermediate! Lets not pretend that this is anything but nonsense and should have been corrected ages ago.
    Let’s stay factual.
    Both Loschbour and Cheddar man did not have the C copy of rs1800414. So no light skin variant for this specific East Asian mutation.
    Just focus on real academic sources.
    The pigmentation of East Asians is dependent on lot of things. So you can not compare a Mesolithic European with a modern Han Chinese who has an umbrella of light skin mutations which is typical in China.
    So when you observe the hirisplex of a modern Chinese, the outcome will NEVER ever be black skinned such as the Cheddar man. So I am not sure why we want to compare the Cheddar man who has practically nihil skin depigmentation with the pigmentation genetics of modern East Asians such as the quite light skinned Chinese people.

    You can find the real Loschbour hirisplex snps here and compare him with the snos of the much lighter skinned Anatolian farmers:
    https://www.biorxiv.org/content/bior...?download=true


    You can check your own hirisplex prediction with the raw data of your own dna tests.
    I have seen several hirisplex predictions from different ethnicities including jews and Middle Easterners. Many people observe them as dark skinned people, socially.
    I can tell I have never seen such a black skin prediction on any modern Europeans, Middle Easterners and East Asians.
    Not even on much darker Austronesians/South East Asians.
    Last edited by Ylang-Ylang; 10-17-2021 at 05:10 PM.

  2. The Following 6 Users Say Thank You to Ylang-Ylang For This Useful Post:

     beyoku (10-17-2021),  Dehlisandwich (10-17-2021),  Hando (10-18-2021),  newbiepunjabi (10-18-2021),  peloponnesian (10-18-2021),  pmokeefe (10-17-2021)

  3. #12
    Because my write up is extensive I'm going to break down some issues here:

    * Single SNP shifts make a large difference especially in Cheddar Man. This means there's very poor fault tolerance. With Cheddar Man it's like it reaches the point where insignificant bits become significant bits and a single bit flip fault has a very high chance of also flipping the whole result.
    * The manual for the tool gives a false sense of security saying all other SNPs are in the correct orientation likely leading to significant cases of user error.
    * People getting the SNPs in the wrong orientation is an acknowledged problem and broad across genetics. Anyone who has jumped into this from another field is very likely to be caught out. The scientific literature is a large dependency graph with the potential for small errors propagating without sufficient redundancy.
    * Cheddar Man really only appears to be differentiated from Europeans by three SNPs. People think these split Europeans from Africans. Those three SNPs are global and it just splits Europeans or rather west Eurasians from everyone else. The tool appears to struggle telling a mix of Europe and African from West and East Eurasian.
    * No appears to have tested these models or trained them for making predictions about unknown groups from known groups. Example, removing all people from one ancestry from the model and training database then seeing what it does when you test it with people from that ancestry. Also not testing if adding that ancestry incorrectly pulls people over from existing groups into it.
    * These kinds of tools appear to be just mapping statistical association rather than causal. It's obvious to everyone that it's certain to be missing key SNPs seen in literature.
    * Many results are completely unverifiable and very little has been done broadly to compensate for this. No one seems to have been trying to predict for offspring results or anything like that to put it through its paces.
    * I can't see any robust tests to instead of working out exactly what the mutations were doing as they stand, which direction they were going in.
    * People using methods like HIrisPlex-S should really always be doing something like testing them with half known samples in addition to known samples to always be able to factor in an error rate that can be established rather than falling victim to both the known and unknown error rates. No assumption that the error rate is higher than measured (an error may flip the same proportion of incorrect results to correct results). Some errors may even produce more correct results. Defaults commonly do this.

    There's a bunch of other more nuanced issues and errors coming at this from Computer Science and hitting the learning curve. These issues tend to stand out fairly clearly. I'm seeing often a lot of missing data all over the place and scope for errors like people not including temporal data (age of sample, separated from immigrant populations, etc) in datasets or getting geographic data wrong (not dividing Africa by the Sahara, confusion over what it really Europe and Asia, etc). I see a lot of science still trying to work things out for modern populations which is very incomplete while then moving on to ancient samples perhaps prematurely.

    Two issues are related. When you play around flipping bits you get a result like the tool thinks it's 50/50 whether Cheddar Man is sub-Saharan African or East Eurasian so it basically doesn't seem to know (only that it recognises the non-Euro Northern basin alleles then undecided which non-Euro population to match or ambiguously non-European due to those alleles). That's a strong indication the tool really finds him in an unknown category.

    What I'm saying is that when I interrogate the tool, its input and output, it's actually telling me that for Cheddar Man it doesn't know and to toss a coin. Could be Nigerian, could be Chinese. My extra tests haven't been incorporated into an output filter or user guide. Deliberately make a few user errors with some random bits one at a time and you tend to see this.
    Last edited by John Baker; 10-17-2021 at 07:58 PM.

  4. The Following User Says Thank You to John Baker For This Useful Post:

     Hando (10-18-2021)

  5. #13
    Ylang-Ylang, despite you showing intelligence and awareness, you've used the tool differently and probably wrongly which is another point I'm making, usability sucks. I work a lot in data and computer science, you would be amazed just how much of a problem data entry is and user error resulting from the user being set up to fail.

    Get ten expert users, get ten different results. You see the problem?

    rs312262906 was apparently renamed to rs796296176 which is why you don't have it for example. Something I have neglected to mention. That could even be flipped for all we know. If it is then I guess it makes him a redhead?

    You're making the same mistakes I did and probably researchers too. The tool looks easy to use up front, in reality it's not. It's booby trapped. Deceptively complicated. It's far easier to get user input wrong than right. Most people are going to get it wrong.

    The tool has another hidden page that shows how you should really process Cheddar Man's results:

    https://walshlab.sitehost.iu.edu/hpsconvertsonline.R

    It does not cover for SNP renames. Even applying these fixes produces somewhat questionable results for my DNA sample, having more samples with reliable phenotype data would be nice.

    I've been in this situation a lot in software and my response is that we can't work like this. This is that same situation when making gambling machines, online gaming, data aggregation, AB testing, satellite communications configuration/allocation/data, general data collection, tracking, accounting, auditing, etc (I have had to fix all manner of software pertaining to data with human interaction and constant nightmare scenarios like this) and I reach that point where I'm saying this is made more difficult than it has to be, it's near impossible for someone, even well versed to get it right.

    You have to assume this situation is a constant. It's just how human work on average and systems tend to evolve. At some point they hit the Gordian Knot threshold from humans disorganising things more as they do more. Common things like using a copy of the data rather than the original and errors creeping in. It doesn't matter what the domain is.

    Ylang-Ylang is also revealing some more missing data or data that's not upfront. The coverage. In the tool 1x coverage is basically NA. I do vaguely remember there being a more expansive spreadsheet document if you have it. That's what I used two years ago but couldn't find it this time around. I remember one of the sheets had all the reads. Cherry picking through papers to rebuild datasets is another problem we'd usually work to eliminate in software development. Usually when I have to rebuild something like that from derived or secondary sources it's because of some major catastrophic data loss disaster. In this case disorganisation appears to be the key contributor.

    matp aren't you also one of the only people to have noticed the ASIP issue two years ago and published it? Something which the press should have easily picked up on but didn't.
    Last edited by John Baker; 10-17-2021 at 07:21 PM.

  6. The Following User Says Thank You to John Baker For This Useful Post:

     Hando (10-18-2021)

  7. #14
    Registered Users
    Posts
    74
    Sex
    Y-DNA (P)
    G-M406
    mtDNA (M)
    M7c1c3

    Quote Originally Posted by Matp View Post
    Yes I have the Loschbour and other WHG snps - the point is to test the consistency of the model. Loschbour and Cheddar man overall have very similar pigmentation snps- and yet get predicted entirely differently on hirisplex-s but their eye predictions are identical. I should add that both get almost identical scores on the online snipper model:

    Attachment 47070Attachment 47071

    I also did a custom database on the snipper with more snps (43 snps), again WHG all get predicted very similar:

    Attachment 47072Attachment 47073

    I have tested its consistency on modern populations and ancient - and the inconsistencies in WHG seem to be on variants most unique to Europe funnily enough. Apparently for WHG, lighter eyes is associated with higher probabilities of black skin...
    It is very easy to experiment with the Cheddar man’s hirisplex and the East Asian OCA2 gene mutation; rs1800414 C

    So let’s see what happens if we change Cheddar man’s rs1800414 TT to rs1800414 CC.
    Even I am shocked to see this huge decrease of pigmentation from dark to black to intermediate!
    This proves that rs1800414 C copy is extremely crucial for the pigmentation of East Asian populations.

    The real Cheddar man Hirisplex outcome:
    CFF02AC9-FE23-4456-BDBC-B7D25CB85682.jpeg

    Here when we modify the Cheddar man’s hirisplex a bit.
    We put 2 for rs1800414.

    Result: exactly 100% intermediate skin prediction, which is very common in East Asia:

    D15516AE-6499-4098-94A8-2DA9105CCE24.jpeg

    P-value
    Intermediate skin 1 (=100%)

    As you see skin pigmentation prediction isn’t that simple and very complicated because a lot of genes are involved.
    So you can not say person A has one light skin copy so she must be fair like the modern Lithuanians.
    Last edited by Ylang-Ylang; 10-18-2021 at 06:11 PM.

  8. The Following 2 Users Say Thank You to Ylang-Ylang For This Useful Post:

     Hando (10-18-2021),  siberoberingian (10-18-2021)

  9. #15
    Registered Users
    Posts
    74
    Sex
    Y-DNA (P)
    G-M406
    mtDNA (M)
    M7c1c3

    The pigmentation of the ancient Aegeans is another interested topic.
    They also explain light skin variants detected in the new study, BUT overal dark skinned predicted. (Not black skinned)
    Dark skinned as the Minoan/Mycenaean men on the ancient wall paintings.
    Many people used to assume it was only fashion, because it is still believed these ancient Greeks must had been ‘pale’ like Scottish people in reality. This has to do with the centuries old bias and fantasy that real ancient Greeks originally must had been a true blonde blue eyed Germanic race and modern Greeks are darkened due to the islamic invasions.
    I do not understand we still get some disbelieving reactions (especially on the internet) because we still see some modern Greeks with darker pigmentation.

    The reality people depicted their own kind just the way they were.
    It was not just a fashionable sunny tan from hours of sunbathing and fake bronzers, it was in the pigmentation genetics of these ancient (BA) Aegean people.

    https://www.sciencedirect.com/scienc...92867421003706

    Phenotypic insights: Pigmentation and lactose intolerance
    Using genotype data, we predicted that Pta08, Kou01, and Log02 most likely had brown eyes, dark brown to black hair, and dark skin (Table S1; STAR Methods). These predictions match the visual representations of male individuals from BA wall paintings of Minoan Crete for hair and eye color. The eye and hair color predictions were similar to those from later periods of the Aegean BA (Lazaridis et al., 2017). Although the overall prediction for all three individuals was of dark skin, they also all carried alleles strongly associated with lighter skin color (rs1426654 in the gene SLC24A5, and rs16891982 in SLC45A2) (Mathieson et al., 2015). The latter is in line with observations that skin depigmentation has been segregating since the Neolithic in southern Europe (Hofmanová et al., 2016; Mathieson et al., 2015).

  10. The Following 2 Users Say Thank You to Ylang-Ylang For This Useful Post:

     Hando (10-18-2021),  peloponnesian (10-18-2021)

  11. #16
    > This proves that rs1800414 C copy is extremely crucial for the pigmentation of East Asian populations.

    I would be careful with that. It may be crucial for the tool to differentiate between those. It may not be what differentiates their pigmentation. Now you've played with the tool a bit you will start to see issues.

    Put in my SNPs from the CSV in the Quora post (left side of Fix4).

    Play with some of the ones that make a big difference for Cheddar Man and see they have no real impact for mine. Then play with either one of multiple Cheddar Man has, they all have a near universal amplified difference. In mine it's noise. So I think what you're seeing is some artefacts of whatever mathematical formula they're using and results that go down a path in the tree of more ambiguity like Cheddar Man starts to amplify noise. Those SNPs make something like 1% or less difference to me but all amplified for Cheddar Man.

    When you do this you get the coin toss effect I mention. The tool doesn't really know as far as I can see. It's really like all SNPs with a tiny amount of noise become critical not far from 50/50 depending on the noise which direction they take it. To me that looks like it's one bit reduced to East Asian or African or rather the tool doesn't know and is having to use the least amount of information, one bit to decide.

    I think your OCA bits are wrong too, or mine is, I'll have to check. It looks like yours.

    Basically, the tool is saying Cheddar Man is Odo, Unknown Sample, a shapeshifter or a changling.

    Another question, what if people could change their pigmentation at will once or were striped? We could clone Cheddar Man and find he was a chameleon long lost genes to change your skin colour according to the population you're in. I would genetically engineer that if I were a mad scientist. Would be cool.

    Believe it or not there's actually a strong scientific basis for transgenerational changlings. Evolution needs the presentation of both outcomes in a gene pool to select against one. Though that may not truly be the case here. Though nature will also produce genetic models with ambiguous or variable outcomes. Usually in evolution we expect models to be destabilised then stabilised through selection. A similar process is happening with this tool though I think that's actually in the tool itself and users, etc.

    When I go through all of CM's alleles that he has but I don't, they're very few and all are present across Eurasian in sufficient frequencies at least according to all of the data I can find, which is academic (though very messy to sort through or verify). So I'd pin him as more toward Eurasian than African. However this obviously needs far more scientific enquiry to really be able to say anything for certain. I tend to be drawn to want to look more at Indian populations and Andaman ancient or Australian ancient.

    In terms of ancestry, Cheddar Man can be considered almost certainly extinct at least in the scope of this gene segments given the presence of SLC genes so really he's probably nothing to do with anyone. A lot of specimens they bring back actually turn out to be extinct humans in Europe. They tie them to modern Europeans which can be misleading. The extinction of those genes suggests some impact which may or may not be pigment. These genes tend to effect a few things and play out in strange ways with others.
    Last edited by John Baker; 10-17-2021 at 09:26 PM.

  12. The Following User Says Thank You to John Baker For This Useful Post:

     Hando (10-18-2021)

  13. #17
    Registered Users
    Posts
    207
    Sex

    Quote Originally Posted by Ylang-Ylang View Post
    Dark skinned as the Minoan/Mycenaean men on the ancient wall paintings. ... I do not understand we still get some disbelieving reactions (especially on the internet) because we still see some modern Greeks with darker pigmentation. ... The reality people depicted their own kind just the way they were.
    You're leaving out the fact that the Minoan women are depicted as pale, bone white.

    Quote Originally Posted by Ylang-Ylang View Post
    it is still believed these ancient Greeks must had been ‘pale’ like Scottish people in reality. This has to do with the centuries old bias and fantasy that real ancient Greeks originally must had been a true blonde blue eyed Germanic race and modern Greeks are darkened due to the islamic invasions.
    Minoans and ancient Greeks aren't the same thing ... There are depictions of blonde and blue eyed ancient Greeks, but not blonde blue-eyed Minoans. According to their DNA a minority of Minoans had blue eyes but their hair seems to have been uniformly dark.
    Last edited by Philjames; 10-17-2021 at 08:21 PM.

  14. The Following 2 Users Say Thank You to Philjames For This Useful Post:

     Ahuwarhd (10-19-2021),  Hando (10-19-2021)

  15. #18
    Registered Users
    Posts
    342
    Sex
    Location
    Canada
    Ethnicity
    Somali
    Y-DNA (P)
    T-L208
    mtDNA (M)
    N1b2

    Canada Somaliland
    Cheddar Man really doesn't seem to be very dark skinned if you pop his data into the skin classifer . The multinomial logistic regression classifier is the only one that seems to give him dark skin.

    Most of the WHG also don't seem to be very dark skinned

    Cheddar Man: GGCCGGGGAAGGAACCAACC Predicted admixture: 62.49 % for White; 37.50 % for Intermediate; 0.01 % for Black.

    Luxembourg_Loschbour: GGNNGGGGAANNAGCTAANN 85.73 % for White; 14.27 % for Intermediate; 0.00 % for Black.

    Bichon: GGCCAAAGAGGGGACCAGCG Predicted admixture: 52.36 % for White; 47.62 % for Intermediate; 0.02 % for Black.

    Chan: GGNNAAGGAANNAACCAANN 93.23 % for Intermediate; 5.55 % for White; 1.22 % for Black.

    LaBrana: GGNNGGGGAANNAACCAANN Predicted admixture: 52.36 % for White; 47.62 % for Intermediate; 0.02 % for Black.


    Although they do seem to have higher odds for dark skin relative to Kotias and Sidelkino

    Kotias: GGNNAAAAAANNAACCAANN Predicted admixture: 99.89 % for White; 0.11 % for Intermediate; 0.00 % for Black.

    Sidelkino: GGGGGGAANNCCAACCGGCC Predicted admixture: 99.98 % for White; 0.02 % for Intermediate; 0.00 % for Black.



    Mota does seem to get an intermediate tone so this predictor may not be the most accurate.

    Mota: GGNNAAGGAANNAACCGGNN Predicted admixture: 89.40 % for Intermediate; 10.54 % for Black; 0.05 % for White.


    This site has a larger selection of alleles that affect phenotype which may give more accurate predictions.

  16. The Following 8 Users Say Thank You to Mnemonics For This Useful Post:

     Dehlisandwich (10-17-2021),  diini95 (10-18-2021),  Hando (10-19-2021),  John Baker (10-18-2021),  Keneki20 (10-18-2021),  laltota (10-18-2021),  theplayer (10-17-2021),  whynot (10-18-2021)

  17. #19
    I'll have to play with these tools.

    I have sequencing from Ancestry, 23andme and Nebula. Nebula is full sequencing with fair coverage hence being able to play with the Hirisplex tool.

    I've not done anything yet like compare all the data files to get some hint of an error rate. Obviously though my DNA is going to be in better condition than Cheddar Man's. The tool pegging me as a blue eyed blond half black half white is what gave the game away.

    You know I think science needs new protocols to store bones and teeth in more cold settings or something to preserve the DNA. Before they thought there was nothing to preserve. I guess though what difference does a few more years make of not being frozen. I'd at least have a protocol not to expose them to a high temperature range than they would have been from where extracted.

    There's a trick you can do with the tools to at least get an idea of how certain it is. Scramble one of each SNP, then a random selection of two, three, etc with some way of dividing the proportion to see how wobbly the result is. This doesn't entirely reveal errors but comparing it can give some sense of whether or nor the tool might be struggling or if they hinge on one thing as a single point of failure.
    Last edited by John Baker; 10-18-2021 at 01:24 AM.

  18. The Following 2 Users Say Thank You to John Baker For This Useful Post:

     Hando (10-19-2021),  Mnemonics (10-18-2021)

  19. #20
    Registered Users
    Posts
    342
    Sex
    Location
    Canada
    Ethnicity
    Somali
    Y-DNA (P)
    T-L208
    mtDNA (M)
    N1b2

    Canada Somaliland
    Quote Originally Posted by John Baker View Post
    I'll have to play with these tools.

    I have sequencing from Ancestry, 23andme and Nebula. Nebula is full sequencing with fair coverage hence being able to play with the Hirisplex tool.

    I've not done anything yet like compare all the data files to get some hint of an error rate. Obviously though my DNA is going to be in better condition than Cheddar Man's. The tool pegging me as a blue eyed blond half black half white is what gave the game away.

    You know I think science needs new protocols to store bones and teeth in more cold settings or something to preserve the DNA. Before they thought there was nothing to preserve. I guess though what difference does a few more years make of not being frozen. I'd at least have a protocol not to expose them to a high temperature range than they would have been from where extracted.

    There's a trick you can do with the tools to at least get an idea of how certain it is. Scramble one of each SNP, then a random selection of two, three, etc with some way of dividing the proportion to see how wobbly the result is. This doesn't entirely reveal errors but comparing it can give some sense of whether or nor the tool might be struggling or if they hinge on one thing as a single point of failure.

    DNA damage, contamination, errors, and the lack of heterozygosity (the vast majority of ancient samples we have are ascertained for a single allele) generally make trying to predict phenotype of a specific ancient sample pretty hard. You can generally work around that with a large enough sample size from a specific population but you'll still need a pretty high coverage diploid sample to be relatively confident about what that specific individual looked like.


    Paleo Eurasian samples

    UstIshim.DG: Predicted admixture: 64.60 % for Intermediate; 35.08 % for Black; 0.32 % for White.

    ZlatyKun.SG: Predicted admixture: 99.71 % for Intermediate; 0.28 % for Black; 0.01 % for White.

  20. The Following User Says Thank You to Mnemonics For This Useful Post:

     Hando (10-19-2021)

Page 2 of 5 FirstFirst 1234 ... LastLast

Similar Threads

  1. Blue-eyed, dark-skinned, earliest modern Briton.
    By JohnHowellsTyrfro in forum Ancient (aDNA)
    Replies: 717
    Last Post: 02-19-2018, 10:08 PM
  2. "No one could see the color blue until modern times"
    By MikeWhalen in forum Linguistics
    Replies: 19
    Last Post: 10-30-2017, 08:14 PM
  3. "Black Finns" in MDLP K16
    By Tomenable in forum Autosomal (auDNA)
    Replies: 15
    Last Post: 05-04-2017, 09:44 PM
  4. Replies: 15
    Last Post: 01-30-2017, 09:51 PM
  5. "Pierce Brosnan gene" - dark hair, blue eyes and freckles
    By avalon in forum Anatomy and Physiology
    Replies: 24
    Last Post: 01-02-2016, 04:32 AM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •