Page 8 of 15 FirstFirst ... 678910 ... LastLast
Results 71 to 80 of 144

Thread: Presence of Peninsular Arab in Southern Europeans

  1. #71
    Registered Users
    Posts
    362
    Sex
    Ethnicity
    Greek
    Nationality
    Australian
    Y-DNA (P)
    J2a
    mtDNA (M)
    J1

    Australia Greece Byzantine Empire
    Quote Originally Posted by Sokoski View Post
    That is interesting that Poland falls under D5a2 in the Mrs a line. I’m not obsessed about it, just curious how it is reflected in my Dna and what not. My grandfather is part Sephardic Jewish/North African-hence why I have been searching for it rather than leaving it as a cluster with Southern Italians.

    My dad’s paternal line is polish. His maternal is more mixed but there are distant ancestors from Bashkortostan and Tuva in Central Russia, so I was wondering if that correlated. Overall, it’s appearing to be too much of a hassle to figure out.
    What matters for the mtDNA haplogroup is the identity of your most distant known ancestor of your mother's mother's mother's ... mother. If she's Polish, I'm just saying there are other Polish people with it apparently. Not that it's common in Poland, by any means. If you test his mtDNA with FTDNA you could see if it's the same cluster or a different one. If it's the same cluster, you could try to work out a common link if possible. High chance that won't be possible though.

    As for Sephardic Jewish, you're not gonna be able to tell that apart with autosomal oracles and even the commercial tests are pretty bad at picking it up. I've mentioned this a few times, but the only thing that matters in the end is your relative matches. Do you match a bunch of Sephardic Jews on some segment of DNA? If so, then you could share a common ancestor with them. You'd then need to study their trees to work if/what the common connection is. That's really the only way to verify it, and yes it's not going to be easy. It's more like detective work.
    Code:
    23abc_AncestryDNA_scaled,0.110408,0.151314,-0.0290383,-0.0507112,0.0018465,-0.0156179,-0.00305514,-0.00138456,-0.00899905,0.00911181,0.00243583,-0.00149867,-0.00431116,0.00344057,-0.00773606,0.00106072,0.00195576,0.00152026,0.00251396,-0.00550264,-0.00786113,-0.00197844,0.0025882,0.00168699,0.000957998

  2. The Following User Says Thank You to 23abc For This Useful Post:

     Seabass (03-04-2021)

  3. #72
    Registered Users
    Posts
    99
    Sex
    Location
    Florida
    Ethnicity
    It/Pl/Sw/NA
    Nationality
    American
    mtDNA (P)
    EV13

    United States Betsy Ross Italy Sicily Poland Russia Tatarstan Sweden
    Quote Originally Posted by 23abc View Post
    What matters for the mtDNA haplogroup is the identity of your most distant known ancestor of your mother's mother's mother's ... mother. If she's Polish, I'm just saying there are other Polish people with it apparently. Not that it's common in Poland, by any means. If you test his mtDNA with FTDNA you could see if it's the same cluster or a different one. If it's the same cluster, you could try to work out a common link if possible. High chance that won't be possible though.

    As for Sephardic Jewish, you're not gonna be able to tell that apart with autosomal oracles and even the commercial tests are pretty bad at picking it up. I've mentioned this a few times, but the only thing that matters in the end is your relative matches. Do you match a bunch of Sephardic Jews on some segment of DNA? If so, then you could share a common ancestor with them. You'd then need to study their trees to work if/what the common connection is. That's really the only way to verify it, and yes it's not going to be easy. It's more like detective work.
    Ah, that makes sense my dad’s mother’s size matches a lot of East Russian and some north chinese so I guess that makes sense? My moms mothers side is from Northern Sweden.

    I do match a bunch of Sephardic Jews in Israel, and my grandpas Dna comes back as half Italian/a quarter Sephardic. It seems difficult to tell the distance between Southern Europeans and Jews though. I’m so happy with all the support on here! It is a lot of information that is helping me piece it all together
    Ancestors locations:

    Maternal-Reggio Calabria, Italy;Potenza, Italy; Orebro, Sweden; Perth, Scotland

    Paternal-Rzeszow, Poland; Dagestan, Russia; Canary Islands, Spain; Provence, France

  4. #73
    Registered Users
    Posts
    904
    Sex
    Location
    Barcelona
    Ethnicity
    Iberian
    Nationality
    Espaņa
    Y-DNA (P)
    E1b1a1-Z6015
    mtDNA (M)
    V

    Spain Badajoz Spain Extremadura
    Quote Originally Posted by Ajeje Brazorf View Post
    Why should there be a "last word"? If you want, you could answer me or at least explain more clearly why my models are "failed". Then I am neither an expert nor a mathematician and I just say that the scaled coordinates are better in my opinion, but also your opinion is very useful for me to understand more.
    I have translated it from Portuguese to English with the translator (the direct link is in the quote), but here you have a criticism of scaled models, it basically says that scaling is not scientifically valid since they are not used. Scaled data makes you lose information, that is why they are less noisy but that does not make them more accurate, quite the opposite.

    Quote Originally Posted by Ruderico View Post
    Simple, because they use essentially the first dimensions, where most of the information is allocated, the upper dimensions are less informative and therefore the noise interferes more. This is even more critical in older samples, such as taforalt. The problem is that they continue to have 25 dimensions, and do not solve the problem for which they were created, eliminate the noise of the models, and they do this with data distortion and loss of information. The permission even has an error, is that nobody uses Euclidean distances in a PCA to measure genetic distances. As Anglesqueville even showed, if that were the goal, it would even be better to totally eliminate the dimensions greater than 8, and keep the rest. More seriously, the methodology would never pass a peer review. If we are not guided by scientifically accredited methods,

    As for the results to be more reliable, I totally disagree, just look at the results of Pedro Ruben's father to see that the escalated coordinates give him between 20% to 25% North African, when in reality the average value in Portugal is around 10/11 % and the national variation is small. If this is more reliable then I don't know what will be less.

    PS: I have never seen a single PCA in genetic studies applying anything minimally similar, what they all do is show PC1-PC2 and occasionally one or two dimensions and that's it.
    Quote Originally Posted by Ruderico View Post
    Yes, there is no Han or Jomon in Portugal (in most cases, of course) but my girlfriend also always gets 2% of that in scaled data. As I said, scaling does not solve the noise problem, there are still 25 dimensions. Whether or not she keeps regular data I don't know why I never use these references, I find it a waste of time to put components with no historical value or genetic evidence as a source in the models, they will only hinder and cause overfitting - which is one of the other problems that the models have and that almost nobody cares about - and ruin the model. I always make custom models for the story of the sample in question, I do not put everything to noise to see what comes out.

    I also don't care about the "basic models" of the G25 and I never do them. I think the samples are too old for this, the vast majority of the higher dimensions are not doing anything there. If they can already cause noise in modern samples - which were used to build the 25 dimensions of the G25 - in samples with 7000 years old that had different variations this is even worse. Climbers may perhaps work better here because of this, but the methodological problem remains.

    It is true that some Euclidean distances from the PCA are strange and difficult / impossible to explain, but no one uses them to measure anything, nor does nMonte use them in the construction of models - this only makes calculations between the coordinates of the different points and reaching a value close to the coordinates of the point being tested. Truth be told, a PCA is a data visualization tool, since applying a Monte Carlo method is making your nose twist a little. With data changed after the PCA data calculation is more dubious.
    I remain in mine: I only make models with sub-modern populations in historical context, using PCA data that was obtained directly from the raw data. Anyone who wants to use others is at ease and I will not criticize (even because occasionally I also run these models that appear there just out of curiosity), but no one has the moral to criticize the scientific honesty behind the regular data, because if they have problems then the climbers will also have, and that would be a problem with the G25 itself. And this one even exists, but it doesn't seem critical to me, the G25 serves to have a good idea of ​​what's going on.

    [1] "distance%=1.9178"

    JJJ

    NW_Iberia_IA,79.8
    Roman_Colonial,12.4
    Berber_EMA,7.8

  5. The Following User Says Thank You to JJJ For This Useful Post:

     Sokoski (03-03-2021)

  6. #74
    Registered Users
    Posts
    944
    Sex

    Quote Originally Posted by JJJ View Post
    I have translated it from Portuguese to English with the translator (the direct link is in the quote), but here you have a criticism of scaled models, it basically says that scaling is not scientifically valid since they are not used. Scaled data makes you lose information, that is why they are less noisy but that does not make them more accurate, quite the opposite.
    - In unscaled coordinates eliminating dimensions larger than 8 does not completely solve the problem and the noise remains.
    - If that user came up with such a proportion (I doubt it) it depends on the samples that are used, this applies to both scaled and unscaled coordinates.
    - Neither Han nor Jomon comes out in my models, actually she only gets Taforalt (6.6 - 6.8%) and Ganj_Dareh (2.4 - 3.6%)
    - In the end Global25 while a great tool, it's still amateur stuff so I have little interest in being scientific. I use what makes the most sense.
    [1] "distance%=2.9299"

    Ajeje Brazorf

    EEF,51.4
    BALTIC_BA,12.6
    GANJ_DAREH,10.8
    CHG,10.4
    MIDDLE_EAST,10
    ANE,2.4
    TAFORALT,1.8
    WHG,0.6

  7. The Following 2 Users Say Thank You to Ajeje Brazorf For This Useful Post:

     23abc (03-03-2021),  Sokoski (03-03-2021)

  8. #75
    Registered Users
    Posts
    362
    Sex
    Ethnicity
    Greek
    Nationality
    Australian
    Y-DNA (P)
    J2a
    mtDNA (M)
    J1

    Australia Greece Byzantine Empire
    Quote Originally Posted by JJJ View Post
    I have translated it from Portuguese to English with the translator (the direct link is in the quote), but here you have a criticism of scaled models, it basically says that scaling is not scientifically valid since they are not used. Scaled data makes you lose information, that is why they are less noisy but that does not make them more accurate, quite the opposite.
    Here is unscaled vs. scaled for myself:

    Code:
    Target: 23abc_AncestryDNA_scaled
    Distance: 1.6776% / 0.01677575
    62.8	Greek_Kos
    17.8	Greek_Dodecanese
    16.0	Greek_Cappadocia
    3.4	Greek_Trabzon
    Code:
    Target: 23abc_AncestryDNA_unscaled
    Distance: 0.8554% / 0.00855389
    45.4	Greek_Dodecanese
    24.6	Greek_Kos
    21.2	Turkish_Kayseri
    7.4	Turkish_Balikesir
    1.4	Turkish_Istanbul
    FYI on paper I am 7/8 Dodecanese + 1/8 Unknown Greek from Constantinopolis. Most likely Unknown was mostly of Anatolian (Cappadocian/Pontic) ancestry, although I never got a relative match related to him on any commercial site.

    So I'm expecting to score roughly 87.5% Dodecanese/Kos + 12.5% Cappadocian/Trabzon.

    Scaled shows 80.6% Dodecanese/Kos + 19.4% Cappadocian/Trabzon, a very respectable estimation.

    Unscaled shows 70% Dodecanese/Kos + 30% Turkish, a very bad estimation. Even though Turks from Kayseri are majority of Cappadocian ancestry, they still have East Eurasian admixture which I lack, so I shouldn't be scoring it in the presence of actual Cappadocian Greek samples.

    And here for my mother:

    Code:
    Target: 23abc_Mother_scaled
    Distance: 2.6520% / 0.02651964
    54.2	Greek_Dodecanese
    32.8	Greek_Cappadocia
    11.2	Turkish_Istanbul
    1.8	Greek_Trabzon
    Code:
    Target: Mother_unscaled
    Distance: 1.8026% / 0.01802633
    61.0	Greek_Cappadocia
    25.6	Greek_Dodecanese
    13.4	Turkish_Balikesir
    My mother on paper is 3/4 Dodecanese and 1/4 Unknown Greek from Constantinopolis.

    So she's expected 75% Dodecanese/Kos + 25% Cappadocian/Trabzon.

    Scaled shows 54.2% Dodecanese/Kos + 34.6% Cappadocian/Trabzon + 11.2% Istanbul Turkish. The Anatolian is a bit high but it's possible some ancestors presumed to be from the Dodecanese were actually also Anatolian, as she doesn't have many close relatives on any commercial website which is the opposite of my father's.

    Unscaled shows 61% Cappadocian/Trabzon 25.6% Dodecanese/Kos + 13.5% West Turkish. This is much more off as like I said most of my mother's ancestors are from the Dodecanese, not Central Anatolia or Pontus.

    In both cases, reference sheet having the same references, the scaled nMonte model not only picked more appropriate choices but more appropriate percentages in both cases compared to the unscaled one. And I see the same thing for nearly everything I have modelled. Maybe others have had different experience, but I can't say I've ever seen nMonte work well with unscaled coordinates.
    Code:
    23abc_AncestryDNA_scaled,0.110408,0.151314,-0.0290383,-0.0507112,0.0018465,-0.0156179,-0.00305514,-0.00138456,-0.00899905,0.00911181,0.00243583,-0.00149867,-0.00431116,0.00344057,-0.00773606,0.00106072,0.00195576,0.00152026,0.00251396,-0.00550264,-0.00786113,-0.00197844,0.0025882,0.00168699,0.000957998

  9. The Following 4 Users Say Thank You to 23abc For This Useful Post:

     Ajeje Brazorf (03-03-2021),  Andrewid (03-04-2021),  bovefex (03-04-2021),  Seabass (03-04-2021)

  10. #76
    Gold Class Member
    Posts
    3,146
    Sex
    Location
    Pennsylvania
    Ethnicity
    West European
    Nationality
    USA
    Y-DNA (P)
    R1b-U152-Z36+FGC6511
    mtDNA (M)
    H11a2a

    United States of America Germany England France Scotland Ireland
    Quote Originally Posted by Ajeje Brazorf View Post
    Why should there be a "last word"? If you want, you could answer me or at least explain more clearly why my models are "failed". Then I am neither an expert nor a mathematician and I just say that the scaled coordinates are better in my opinion, but also your opinion is very useful for me to understand more.
    Very well, then. To answer the questions:

    Leaving aside the talk about the models The models are one of the primary reasons I am using the G25. Specifically, I want to be able to do modern models for mixed individuals to check for plausibility of paper ancestry combined with new things learned from autosomal research. Admixture, including overall and targeted, combined with matching "cousins" is important for this. So, it is not a side issue for me. It really is my primary interest.

    How come the distances are getting weirder with unscaled coordinates? I am not sure how you generated such a list, but this isn't my experience. To represent, here are my mom's matches. She is a little over half South German/West German/French and a little under half Isles, with some minor Amerindian, SSA, and some persistent minor South Asian with a Balkan signal too that led in my research to mixed Roma "cousins" through segment searching, triangulation, etc. These are the unscaled matches:

    Welsh:WalesDR56 - 1.345%
    Afrikaner:AFR057 - 1.449%
    English:HG01790 - 1.530%
    French_Brittany:French24061 - 1.535%
    French_Nord:N_52 - 1.549%
    Swiss_German:Swiss_German5 - 1.569%
    Afrikaner:AFR049 - 1.608%
    French_Nord:N_19 - 1.615%
    French_Brittany:French24090 - 1.623%
    French_Auvergne:C_82 - 1.665%
    French_Alsace:A_41_2 - 1.672%
    Afrikaner:AFR053 - 1.691%
    Afrikaner:AFR018 - 1.696%
    Welsh:WalesL40 - 1.705%
    French_Auvergne:C_16 - 1.722%
    Afrikaner:AFR014 - 1.723%
    French_Brittany:Rennes_B_20 - 1.725%
    French_Occitanie:T_70 - 1.726%
    English_Cornwall:HG00259 - 1.729%
    Irish:Irish29 - 1.730%

    These are the scaled:

    French_Brittany:French24061 - 2.267%
    Welsh:WalesDR56 - 2.289%
    French_Brittany:Rennes_B_29 - 2.460%
    French_Nord:N_52 - 2.519%
    French_Brittany:French23830 - 2.522%
    Welsh:WalesL40 - 2.595%
    English:HG01790 - 2.608%
    English_Cornwall:HG00235 - 2.612%
    Afrikaner:AFR053 - 2.620%
    French_Brittany:Rennes_B_20 - 2.646%
    English:HG01791 - 2.669%
    English_Cornwall:HG00259 - 2.707%
    Afrikaner:AFR057 - 2.717%
    French_Brittany:French24090 - 2.755%
    English_Cornwall:HG00255 - 2.760%
    French_Occitanie:T_70 - 2.781%
    French_Brittany:Rennes_B_46 - 2.791%
    Irish:Irish46 - 2.835%
    French_Brittany:Rennes_B_6 - 2.841%
    Orcadian:HGDP00805 - 2.845%

    There isn't enough difference in the nature of these lists in my opinion to convince me to exclude or minimize by scaling, exaggerating other dimensions and thereby distorting the original PCA coordinates for my modeling.

    Why is it that unscaled hierarchical clustering also seem less consistent than scaled hierarchical clustering? I spent a short amount of time looking at the two different hierarchies. I don't see anything egregious in either one. I have a simliar statement on this: There isn't enough difference in the nature of these lists in my opinion to convince me to exclude or minimize by scaling, exaggerating other dimensions and thereby distorting the original PCA coordinates for my modeling.

    If you want, you could answer me or at least explain more clearly why my models are "failed". My experience with the unscaled modeling with modern individuals from the standard G25 spreadsheet and using Gradient Descent in my Ancestry calculator https://yk.github.io/ancestry/ is that once the distance starts to get too far above 0.95%, something is missing and as the models start to get lower than 0.70% they often start to overfit producing unwarranted results. The lower number especially can have some variability in the ideal number if one is really close to some of the references, but generally I see for a mixed individual that there is a nice spot around 0.80% to 0.85% more or less a little is often a plausible model. Saying that "unscaled" modeling doesn't work while presenting a model that has a distance of 1.50% isn't very convincing. Neither is presenting a model with several overlapping reference groups and expecting the tool to sort this out. Remember that I am talking modeling with unscaled, modern individuals.

    If I could demonstrate with my mom's kit, I likely can make this plainer. If I model her unscaled with just her majority ancestry (picking some proxy reference groups), I get something like this:

    Distance: 0.884%

    Welsh: 40.9%
    Swiss_German: 39.4%
    French_Paris: 19.7%

    It is a nice safe, model, accurate, and supported by my paper ancestry research. It says my paper ancestry research is "plausible". If I do the same model with "scaled" references, I get this:

    Distance: 1.540%

    Welsh: 51.0%
    Swiss_German: 34.0%
    French_Paris: 15.0%

    This isn't bad. It is less accurate, as it thinks the Isles ancestry is more prominent, when it isn't in reality. But, I don't hate it. It is close enough, but I think the other one is better. If I want to follow the minor ancestry signals to where they seem to have gone when I follow their trails into my mother's Appalachian colonial USA ancestry where she triangulates with cousin's that have the same admixture in the same places, but they have "more" to help with my model as they indicate the source of the admixture more clearly, I can try to test for "plausibility":

    Distance: 0.842%

    Welsh: 40.6%
    Swiss_German: 34.3%
    French_Paris: 20.8%
    Roma_Balkans: 2.8%
    Mbuti: 0.8%
    Cree: 0.8%

    This is good. It matches admixture tests, except for the African is a bit higher than admixture says it is, but it still is a small number. If I try the same plausibility test with scaled, I get this:

    Distance: 1.534%

    Welsh: 50.6%
    Swiss_German: 34.8%
    French_Paris: 14.5%
    Mbuti: 0.2%

    This isn't much help. The problem is that the scaling minimizes too much information on its quest to destroy "noise". I would rather fight noise by doing some legwork and creating a plausible model. Some of those dimensions of the PCA that are reduced to almost no effect by scaling do things like distinguish among different kinds of Asians or Africans, etc. I would rather have that information passed to the optimizer than render it useless. So, if I scale, I exaggerate to the optimizer dimensions like PC1 and PC2 where it is hard to tell some of my family one from the other (really, we are close genetically, so it isn't completely inaccurate to do that), see the plots here:

    https://anthrogenica.com/showthread....l=1#post566346

    And, I reduce to little or almost no value dimensions like PC 10 where my family, still looking very European, have some interesting distinctions, see the plots here:

    https://anthrogenica.com/showthread....l=1#post567084

    So, that is it. This is why I am not very interested in scaled modeling. I am sure if one is relatively unmixed or very simply mixed that scaling can reduce the work to produce some kinds of models. It seems to help very ancient modeling, too. But, the scaling throws away too much information that I want the optimizer to consider. Sure, one can make giant mistakes without being careful, but I would rather be careful than not to use the PCA information from the formal output of the process. I still am a "whatever floats your boat, whatever works for you" on this. I think the dogmatic "you really need to use scaled coordinates or you are less than human and against science" is just not very convincing to me.
    Last edited by randwulf; 03-04-2021 at 04:07 AM.

  11. The Following 3 Users Say Thank You to randwulf For This Useful Post:

     JerryS. (03-04-2021),  Sokoski (03-04-2021),  ThaYamamoto (03-04-2021)

  12. #77
    Registered Users
    Posts
    635
    Sex

    Scotland England Australia Star of David Turkey Egypt
    Quote Originally Posted by 23abc View Post
    gh.

    As for Sephardic Jewish, you're not gonna be able to tell that apart with autosomal oracles
    This is true. Almost all Sephardic Jews will share much of the same ancient components found in especially Sicilians and Maltese (just in very different proportions!) when compared with Dodocanese Greeks who lack really anything North African and even likely Latin despite a brief history of Italian/Venetian rule.

    Quote Originally Posted by Sokoski View Post
    It seems difficult to tell the distance between Southern Europeans and Jews though. I’m so happy with all the support on here! It is a lot of information that is helping me piece it all together
    You may find this PCA I posted in the link below of interest. You can see a big distinction between Sephardic Jews and Southern Europeans. Their cline towards West Asia from Southern Europe is somewhat parallel to some Greek Islanders/Cypriots.

    https://anthrogenica.com/showthread....l=1#post732645

  13. The Following 3 Users Say Thank You to Seabass For This Useful Post:

     23abc (03-04-2021),  passenger (03-04-2021),  Sokoski (03-04-2021)

  14. #78
    Registered Users
    Posts
    2,740
    Sex
    Location
    America
    Ethnicity
    North & Ionian Seas
    Nationality
    American
    Y-DNA (P)
    I1 (P109)

    England Italy Germany Scotland
    Quote Originally Posted by Nino90 View Post
    My brother and father "score" some Berber/North African and some Saudi.


    I think it is trace or just calculator effect:















    Here is individual samples:
    Here is individual samples:

    Target: Brother_scaled
    Distance: 0.8883% / 0.00888259
    18.6 Irish
    18.2 German
    17.0 Danish
    12.4 Dutch
    9.8 Saami
    9.0 Swiss_French
    7.4 Spanish_Soria
    2.0 Moroccan_South

    1.8 Welsh
    1.6 BedouinB
    1.0 Karelian
    0.6 Nganasan
    0.4 Adygei
    0.2 Spanish_La_Rioja


    wow, your brother and I look similar with the North of the Alps and points south pops (except for his Sammi like pops). his is Iberian leaning and mine looks Balkan leaning.


    Target: JerryS._scaled
    Distance: 0.3438% / 0.00343819
    23.2 Irish
    21.2 Shetlandic
    14.2 German
    13.2 Swiss_French
    9.6 Dutch
    8.0 Danish
    4.8 Welsh
    3.2 Greek_Laconia
    1.0 Berber_MAR_ERR
    1.0 French_Occitanie
    0.6 Masai


    The full spreadsheet seems to give a north lean.
    Last edited by JerryS.; 03-04-2021 at 11:52 AM.

  15. The Following 2 Users Say Thank You to JerryS. For This Useful Post:

     Nino90 (03-04-2021),  Sokoski (03-05-2021)

  16. #79
    Registered Users
    Posts
    128
    Sex
    Omitted

    Quote Originally Posted by randwulf View Post
    Very well, then. To answer the questions:

    Leaving aside the talk about the models The models are one of the primary reasons I am using the G25. Specifically, I want to be able to do modern models for mixed individuals to check for plausibility of paper ancestry combined with new things learned from autosomal research. Admixture, including overall and targeted, combined with matching "cousins" is important for this. So, it is not a side issue for me. It really is my primary interest.

    [I]
    Welsh:WalesDR56 - 1.345%
    Afrikaner:AFR057 - 1.449%
    English:HG01790 - 1.530%
    French_Brittany:French24061 - 1.535%
    French_Nord:N_52 - 1.549%
    Swiss_German:Swiss_German5 - 1.569%
    Afrikaner:AFR049 - 1.608%
    French_Nord:N_19 - 1.615%
    French_Brittany:French24090 - 1.623%
    French_Auvergne:C_82 - 1.665%
    French_Alsace:A_41_2 - 1.672%
    Afrikaner:AFR053 - 1.691%
    Afrikaner:AFR018 - 1.696%
    Welsh:WalesL40 - 1.705%
    French_Auvergne:C_16 - 1.722%
    Afrikaner:AFR014 - 1.723%
    French_Brittany:Rennes_B_20 - 1.725%
    French_Occitanie:T_70 - 1.726%
    English_Cornwall:HG00259 - 1.729%
    Irish:Irish29 - 1.730%

    These are the scaled:

    French_Brittany:French24061 - 2.267%
    Welsh:WalesDR56 - 2.289%
    French_Brittany:Rennes_B_29 - 2.460%
    French_Nord:N_52 - 2.519%
    French_Brittany:French23830 - 2.522%
    Welsh:WalesL40 - 2.595%
    English:HG01790 - 2.608%
    English_Cornwall:HG00235 - 2.612%
    Afrikaner:AFR053 - 2.620%
    French_Brittany:Rennes_B_20 - 2.646%
    English:HG01791 - 2.669%
    English_Cornwall:HG00259 - 2.707%
    Afrikaner:AFR057 - 2.717%
    French_Brittany:French24090 - 2.755%
    English_Cornwall:HG00255 - 2.760%
    French_Occitanie:T_70 - 2.781%
    French_Brittany:Rennes_B_46 - 2.791%
    Irish:Irish46 - 2.835%
    French_Brittany:Rennes_B_6 - 2.841%
    Orcadian:HGDP00805 - 2.845%

    There isn't enough difference in the nature of these lists in my opinion to convince me to exclude or minimize by scaling, exaggerating other dimensions and thereby distorting the original PCA coordinates for my modeling.

    Why is it that unscaled hierarchical clustering also seem less consistent than scaled hierarchical clustering? I spent a short amount of time looking at the two different hierarchies. I don't see anything egregious in either one. I have a simliar statement on this: There isn't enough difference in the nature of these lists in my opinion to convince me to exclude or minimize by scaling, exaggerating other dimensions and thereby distorting the original PCA coordinates for my modeling.

    If you want, you could answer me or at least explain more clearly why my models are "failed". My experience with the unscaled modeling with modern individuals from the standard G25 spreadsheet and using Gradient Descent in my Ancestry calculator https://yk.github.io/ancestry/ is that once the distance starts to get too far above 0.95%, something is missing and as the models start to get lower than 0.70% they often start to overfit producing unwarranted results. The lower number especially can have some variability in the ideal number if one is really close to some of the references, but generally I see for a mixed individual that there is a nice spot around 0.80% to 0.85% more or less a little is often a plausible model. Saying that "unscaled" modeling doesn't work while presenting a model that has a distance of 1.50% isn't very convincing. Neither is presenting a model with several overlapping reference groups and expecting the tool to sort this out. Remember that I am talking modeling with unscaled, modern individuals.
    Aren't those matches simply distance?

    Also, how is a distance of 1.50% supposed to be bad? From my experience it's sometimes literally impossible to below 2.0% on Vahaduo with 0.25 collision, unless you want to include absolute nonsensical samples, such as those from literal continents away.

  17. #80
    Registered Users
    Posts
    128
    Sex
    Omitted

    Quote Originally Posted by 23abc View Post
    ,but I can't say I've ever seen nMonte work well with unscaled coordinates.
    How are unscaled coordinates for distance vs. scaled?

Page 8 of 15 FirstFirst ... 678910 ... LastLast

Similar Threads

  1. Replies: 5
    Last Post: 10-03-2020, 06:36 AM
  2. Replies: 1
    Last Post: 10-04-2019, 05:09 PM
  3. New dating evidence of the early presence of hominins in Southern Europe
    By paoloferrari in forum Archaeology (Prehistory)
    Replies: 0
    Last Post: 09-11-2017, 02:06 PM
  4. Middle East admixture among Southern Europeans
    By Flavio in forum Ancient (aDNA)
    Replies: 5
    Last Post: 02-28-2017, 07:10 PM
  5. Replies: 43
    Last Post: 01-18-2015, 02:11 PM

Tags for this Thread

Posting Permissions

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