Page 6 of 12 FirstFirst ... 45678 ... LastLast
Results 51 to 60 of 114

Thread: Automated qpGraph: possibilities for the consideration of technically savvy Users?

  1. #51
    Registered Users
    Posts
    609
    Sex
    Ethnicity
    Celto-Germanic, Near East
    Nationality
    Murcan
    Y-DNA (P)
    R1b-U152
    mtDNA (M)
    T2a1a

    United States of America
    Quote Originally Posted by theplayer View Post
    @K33

    Haven't there been several large migrations into america though: Na-Dene-related, "Paleo-Eskimo"-related and Eskimo-Aleut which are all much more recent than 20kya(and likely others, in the other direction as well, we don't know of yet).
    Yes, but Reich lab specifically said they tried all those Siberian/NeoAmerican admixture sources -- Kolyma, Nganasan, Neosiberian, PaleoEskimo etc to improve the fits of Antilles_archaic on their qpgraph. None of them helped

  2. The Following User Says Thank You to K33 For This Useful Post:

     theplayer (01-15-2021)

  3. #52
    Registered Users
    Posts
    1,891
    Sex
    Omitted

    There doesn't appear to be anything unusual about the Antilles_archaic samples. They fit within normal Amerind ranges from what I'm seeing in my F3 outgroup spreadsheet.

    EDIT: Hmm... is there a way to verify the F2 data came out right? I did extract_f2, it managed to extract the afs (I specified a number of populations) but overwhelmed my memory when it went to convert that afs into f2 blocks. I ran afs_to_f2 and it did every pop in my dataset at lightning speed. I guess extract_f2 ignored my specified populations because afs_to_f2 doesn't even deal with the dataset directly, so it would have no way of knowing them. Still seems kind of strange though. Ok I think I see, the snpdat.tsv file is very small, much smaller than when I only ran 12 pops, something went wrong.

    EDIT2: Anyone try using an initgraph with any of the recent updates? Idk if I'm brainfarting and doing something wrong but it's not recognizing any starter graphs
    Error in degree(graph, mode = "in") : Not a graph object
    is_valid does this...
    Error in igraph::degree(graph, mode = "in") : Not a graph object
    Qpgraph however recognizes the graph file. I tested the last two graphs I posted because I wanted to get the worst F4 z-score. The scores came out within a point which seems fine I think the weights are calculated slightly differently by default between qpgraph and find_graphs, worst Z-score on both was 3.70, close to success by that measure. I'm going to try and do a run with opt_worst_residual to see if it can find a model with Z <3 before upping from 9 to 10 admixtures.
    Last edited by Kale; 01-15-2021 at 07:13 PM.
    Collection of 14,000 d-stats: Hidden Content Part 2: Hidden Content Part 3: Hidden Content PM me for d-stats, qpadm, qpgraph, or f3-outgroup nmonte models.

  4. #53
    Registered Users
    Posts
    57
    Sex
    Location
    Sweden
    Nationality
    Swedish

    Sweden Skåne Sweden Kalmar Union European Union United Nations
    Is the graph in igraph format?
    I get that error when trying to use an edgelist where an igraph should be used.

    edges_to_igraph(graph) converts from edgelist to igraph

    Or for some other formats:
    "To load an existing graph, use parse_qpgraph_graphfile() if it’s in the original ADMIXTOOLS format, read_table2() if it’s an edge list, or readRDS() if it was saved in R using saveRDS()."
    Last edited by theplayer; 01-15-2021 at 08:09 PM.

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

     Kale (01-15-2021)

  6. #54
    Registered Users
    Posts
    1,891
    Sex
    Omitted

    Interesting, I'll give that a try. It is in whatever format this is..

    edge B A B
    edge C B C
    admix D B C

    EDIT:
    Here's another from-scratch 9 admixture run of Africa. Score was 63.36, basically the same as others. A few repeating trends...
    - Common South-Africa-related branch that contributes to both Neanderthal and OOA.
    - About 13% Basal Human and 7% Taforalt in Mende
    One interesting difference...
    - ShumLaka gets 6% proto-Taforalt rather than any sort of strange para/basal-Eurasian, I'd say that is a lot more plausible.
     

    digraph G {
    size = "7.5,10";
    South_Africa_2000BPSG_l -> MbutiDG [ label = "7" ];
    LoschbourDG_u -> LoschbourDG [ label = "35" ];
    root -> ChimpREF [ label = "47" ];
    root -> rootw [ label = "47" ];
    Rrr -> DinkaDG [ label = "1" ];
    admixb -> LoschbourDG_u [ label = "10" ];
    admixb -> OngeDG [ label = "37" ];
    Rrlr -> Rrlrh_p [ label = "3" ];
    Rrlrh_p -> Rrlrh [ label = "2" ];
    Rrr -> Rrra [ label = "6" ];
    LoschbourDG_u -> admix [ style=dotted, label = "54%" ];
    Rrra -> admix [ style=dotted, label = "46%" ];
    rootw -> rootwx [ label = "34" ];
    rootwx -> DenisovaDG [ label = "44" ];
    admixi -> NeanderthalDG [ label = "68" ];
    rootwx -> admixi [ style=dotted, label = "82%" ];
    Rrra -> Rrraq [ label = "3" ];
    admixo -> MotaSG [ label = "150" ];
    Rrraq -> admixo [ style=dotted, label = "73%" ];
    rootw -> rootwa [ label = "50" ];
    Rrl -> admixh [ style=dotted, label = "80%" ];
    admixh -> MendeDG [ label = "4" ];
    South_Africa_2000BPSG_l -> South_Africa_2000BPSG_lt [ label = "3" ];
    rootwlk -> admixf [ style=dotted, label = "56%" ];
    Rrlrhy -> admixi [ style=dotted, label = "18%" ];
    Rrraq -> admixiw [ style=dotted, label = "92%" ];
    admixiw -> admixb [ label = "30" ];
    Rrlrhy -> admixiw [ style=dotted, label = "8%" ];
    admixhh -> Taforalt [ label = "35" ];
    rootwa -> admixy [ style=dotted, label = "67%" ];
    admixy -> admixh [ style=dotted, label = "20%" ];
    admixhh -> admixy [ style=dotted, label = "33%" ];
    Rrlrh -> admixo [ style=dotted, label = "27%" ];
    admix -> admixn [ label = "2" ];
    admixn -> admixhh [ label = "18" ];
    admixf -> admixe [ style=dotted, label = "94%" ];
    admixe -> Cameroon_ShumLaka_8000BPDG [ label = "7" ];
    admixn -> admixe [ style=dotted, label = "6%" ];
    South_Africa_2000BPSG_lt -> South_Africa_2000BPSG_ltk [ label = "12" ];
    admixez -> South_Africa_2000BPSG [ label = "20" ];
    South_Africa_2000BPSG_ltk -> Rrl [ label = "1" ];
    South_Africa_2000BPSG_lt -> admixez [ style=dotted, label = "20%" ];
    Rrlrh -> admixez [ style=dotted, label = "80%" ];
    Rrlrh_p -> Rrlrhy [ label = "93" ];
    rootwa -> Rrlr [ label = "9" ];
    Rrlr -> rootwlk [ label = "7" ];
    rootwlk -> South_Africa_2000BPSG_l [ label = "7" ];
    Rrl -> admixf [ style=dotted, label = "44%" ];
    South_Africa_2000BPSG_ltk -> Rrr [ label = "3" ];
    }
    Last edited by Kale; 01-16-2021 at 05:03 AM.
    Collection of 14,000 d-stats: Hidden Content Part 2: Hidden Content Part 3: Hidden Content PM me for d-stats, qpadm, qpgraph, or f3-outgroup nmonte models.

  7. #55
    Registered Users
    Posts
    609
    Sex
    Ethnicity
    Celto-Germanic, Near East
    Nationality
    Murcan
    Y-DNA (P)
    R1b-U152
    mtDNA (M)
    T2a1a

    United States of America
    These results were returned with 4 generations... I went up to 8 generations and it said no better graphs available. Out of 8,281 z-scores only 4 were > |4.00|. 99.5% of the z-scores were <|3|

    This graph has the Ancient Beringian as a nearly 50/50 split between something related to Tianyuan and another clade basal to the much younger Canada_Lucier (Ancient Ontario). It once again shows West Eurasians taking Amerind-related ancestry, this time 17% from a pre-Ancestral Beringian clade.

    I keep telling you guys, humans made it to the Americas much earlier than 20kya. They must have been there when Ust Ishim and Tianyuan were alive--the Beringian land bridge was open 50kya-40kya. There is cave art in Borneo and Mungo Man was in Australia around this same time. If we could only get a pre-Mesolithic Amerind sample... the only reason we haven't yet is because all the archaeologists stop digging as soon as they reach the Clovis layer (combined with frustrating DNA sampling restriction agreements with North American tribes). They need to dig deeper.




    EDIT: upon removing Ancestral Beringian, WHG and Yana still take Amerind-related ancestry-- this time 41% from a ghost clade comprised of 78% Tianyuan/UstKyakhta-related and 22% Canada_Lucier-related ancestry.



    EDIT: and another one, this time with Saqqaq and Natufian added and Hoabhinian swapped for Onge. Yana RHS takes 10% Anzick-related ancestry. Natufians modeled as nearly 50/50 Basal Eurasian/WHG-related as claimed in the original paper.

    Last edited by K33; 01-16-2021 at 05:43 AM.

  8. #56
    Registered Users
    Posts
    1,891
    Sex
    Omitted

    Trying adding another mixture to bring the score and WR down.
    Collection of 14,000 d-stats: Hidden Content Part 2: Hidden Content Part 3: Hidden Content PM me for d-stats, qpadm, qpgraph, or f3-outgroup nmonte models.

  9. The Following User Says Thank You to Kale For This Useful Post:

     Ryukendo (02-04-2021)

  10. #57
    Registered Users
    Posts
    609
    Sex
    Ethnicity
    Celto-Germanic, Near East
    Nationality
    Murcan
    Y-DNA (P)
    R1b-U152
    mtDNA (M)
    T2a1a

    United States of America
    Quote Originally Posted by Kale View Post
    Trying adding another mixture to bring the score and WR down.
    That's one of the negatives to my n00b coding skills, I'm doing everything thru the shiny interface and while I can increase/decrease admixture edges manually under the "Modify" tab, when I run the "Optimize" tab (which is how I've generated nearly all of these final graphs), the admixture control seems to revert back to default.

    Here under the Modify tab I increased the Admix from 2 to 3, then simply removed Antilles_Archaic to "execute" the command. The fit improved, but now we have 28% admixture from Saqqaq into the Tianyuan/Hoabinhian clade. But then it froze up on me when I tried to insert Antilles back into an invalid vertex


  11. #58
    Registered Users
    Posts
    1,891
    Sex
    Omitted

    To do detailed modeling you'll need to use the command line version. AFAIK the slider for # generations on shiny only goes up to 20.
    Also you have to make sure if you are going to include low-coverage samples like Natufian and Hoabinhian that you have maxmiss set above 0 to not trash your snp count.
    Collection of 14,000 d-stats: Hidden Content Part 2: Hidden Content Part 3: Hidden Content PM me for d-stats, qpadm, qpgraph, or f3-outgroup nmonte models.

  12. The Following 2 Users Say Thank You to Kale For This Useful Post:

     K33 (01-16-2021),  Ryukendo (02-04-2021)

  13. #59
    Registered Users
    Posts
    609
    Sex
    Ethnicity
    Celto-Germanic, Near East
    Nationality
    Murcan
    Y-DNA (P)
    R1b-U152
    mtDNA (M)
    T2a1a

    United States of America
    Quote Originally Posted by Kale View Post
    To do detailed modeling you'll need to use the command line version. AFAIK the slider for # generations on shiny only goes up to 20.
    Also you have to make sure if you are going to include low-coverage samples like Natufian and Hoabinhian that you have maxmiss set above 0 to not trash your snp count.
    Yeah I knew Hoabhinians were low coverage, I didn't realize Natufian were so low I just saw they were all like .1 coverage. Thanks for the advice re: maxmiss I'll probably stick with Onge instead

    I noticed that changing the # generations is by far the best thing you can do thru shinytools to improve the fit.... what exactly *is* "generations" though? Does it improve the "depth perception" (for lack of a better word) for sorting thru samples of greatly varying ages... or is it something else?

    Also, is there a one-stop shop I can view the coverage for all the samples in the dataset? Or do I gotta just consult all the various Supp Data files...

  14. #60
    Registered Users
    Posts
    1,891
    Sex
    Omitted

    This has good info on a lot of samples
    https://reichdata.hms.harvard.edu/pu....1240K_HO.anno

    Here's kind of what generations are...
    Generation 0: The software's first attempt of making a graph.
    Generation 1: Takes say 50 instances of Gen0 graph and makes a random change to each and shows you the best one.
    Generation 2: Takes say 50 instances of Gen1's best graph and makes a random change to each and shows you the best one.
    Etc.

    EDIT: Here's a graph of Africa with 10 admixtures... Still a South-African branch mediating the relationship between Neanderthal and OOA.
    No basal Human in Mende this time, but rather ShumLaka, who also gets that weird para/basal Eurasian again.
    Really bizarro around the Taforalt area though.
     

    digraph G {
    size = "7.5,10";
    Rrrlr -> MbutiDG [ label = "6" ];
    LoschbourDG_a -> Rrrlrll [ style=dotted, label = "52%" ];
    LoschbourDG_a -> LoschbourDG [ label = "34" ];
    Rrrll_x -> DenisovaDG [ label = "43" ];
    root -> Rrrll_x [ label = "35" ];
    rooty -> ChimpREF [ label = "47" ];
    rooty -> root [ label = "47" ];
    Rrrlrll_x -> Rrrlrll [ style=dotted, label = "48%" ];
    Rrrll -> Rrrlr [ label = "7" ];
    Rrrll_x -> admix [ style=dotted, label = "79%" ];
    admix -> NeanderthalDG [ label = "69" ];
    Rrrllt -> admixv [ style=dotted, label = "37%" ];
    root -> roott [ label = "53" ];
    Rrrlrl -> admixr [ style=dotted, label = "79%" ];
    admixr -> MendeDG [ label = "3" ];
    roott -> roottd [ label = "6" ];
    Rrrlrl_z -> admixb [ style=dotted, label = "36%" ];
    admixb -> Cameroon_ShumLaka_8000BPDG [ label = "7" ];
    admixs -> LoschbourDG_a [ label = "11" ];
    admixs -> OngeDG [ label = "37" ];
    DinkaDG_lx -> admixv [ style=dotted, label = "63%" ];
    Rrrlrll_x -> DinkaDG_lx [ label = "6" ];
    Rrrlrll -> Rrrlrllh [ label = "29" ];
    Rrrlrllh -> Taforalt [ label = "25" ];
    Rrrlrllh -> admixw [ style=dotted, label = "7%" ];
    Rrrllt -> South_Africa_2000BPSG [ label = "20" ];
    Rrrllv -> Rrrll [ label = "2" ];
    roottd -> admixw [ style=dotted, label = "93%" ];
    rootti -> admix [ style=dotted, label = "21%" ];
    DinkaDG_lx -> admixz [ style=dotted, label = "88%" ];
    rootti -> admixz [ style=dotted, label = "12%" ];
    roottd -> rootti [ label = "62" ];
    roott -> Rrrllv [ label = "13" ];
    admixz -> admixzq [ label = "3" ];
    admixzq -> admixs [ label = "24" ];
    Rrrllv -> admixp [ style=dotted, label = "84%" ];
    admixp -> admixb [ style=dotted, label = "64%" ];
    admixzq -> admixp [ style=dotted, label = "16%" ];
    admixw -> admixwf [ label = "0" ];
    admixwf -> admixr [ style=dotted, label = "21%" ];
    Rrrll -> admixy [ style=dotted, label = "61%" ];
    admixy -> Rrrllt [ label = "4" ];
    admixwf -> admixy [ style=dotted, label = "39%" ];
    admixv -> admixvx [ label = "3" ];
    admixvx -> MotaSG [ label = "148" ];
    admixo -> DinkaDG [ label = "3" ];
    admixvx -> admixo [ style=dotted, label = "29%" ];
    Rrrlrl -> Rrrlrll_x [ label = "8" ];
    Rrrlr -> DinkaDG_l [ label = "20" ];
    DinkaDG_l -> Rrrlrl_z [ label = "0" ];
    Rrrlrl_z -> admixo [ style=dotted, label = "71%" ];
    DinkaDG_l -> Rrrlrl [ label = "1" ];
    }
    Last edited by Kale; 01-16-2021 at 07:29 AM.
    Collection of 14,000 d-stats: Hidden Content Part 2: Hidden Content Part 3: Hidden Content PM me for d-stats, qpadm, qpgraph, or f3-outgroup nmonte models.

Page 6 of 12 FirstFirst ... 45678 ... LastLast

Similar Threads

  1. An American Wellspring: Concepts and Possibilities
    By Alvah Hicks in forum American
    Replies: 2
    Last Post: 09-20-2020, 07:31 PM
  2. Playing around with qpGraph
    By Chad Rohlfsen in forum Ancient (aDNA)
    Replies: 34
    Last Post: 06-19-2017, 10:45 PM
  3. Technically am I part east African or not?
    By jaderose22 in forum African
    Replies: 15
    Last Post: 03-26-2017, 04:59 AM
  4. Replies: 0
    Last Post: 10-04-2016, 12:01 PM
  5. Future possibilities for AncestryDNA
    By Erik in forum AncestryDNA
    Replies: 0
    Last Post: 08-31-2014, 12:35 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
  •