Making two-way G25 models with negative weights

[Nganasankhan]

You lost your avatar. Looks naked. lol

Thanks for this code series.

I have enabled the options to hide avatars and signatures, so I forgot that I even had an avatar. It was the Komi sorcerer Šypiča (https://twitter.com/jurgenfug/status...03139905454080).

I have a library of over 2000 custom functions and aliases in my `.bashrc`, and almost all of them have 1-4 character names, so I can write shell commands in a much shorter than usual form. I have started to make a similar library for R. So for example with these lines from my R configuration files:

library(magrittr)
`|`=`%>%`
`%@%`=`%<>%`

amu=function(x)unname(as.matrix(x))
anu=function(x)unname(as.numeric(x))
ax=function(x,y)sapply(x,eval.parent(call("function",as.pairlist(alist(x=)),substitute(y))))
drn=function(x,...)x[!rownames(x)%in%c(...),]
euo=function(x,y)sqrt(outer(rowSums(x^2),rowSums(y^2),'+')-tcrossprod(x,2*y))
h5=function(x)head(x,2^5)
nr=nrow
o=order
p=function(...)writeLines(as.character(c(...)))
pas=paste
rc3=function(...)read.csv(...,row.names=1)
rn=rownames
rof=function(x,y)sprintf(paste0("%.",y,"f"),x)
spar=function(x,y)formatC(x,y,format="s")
swaw=function(w,x,y){x2=x;y2=y;y2[w]=x[w];x2[w]=y[w];e=parent.frame();do.call("=",list(substitute(x),x2),envir=e);do.call("=",list(substitute(y),y2),envir=e)}
ul=unlist

I can use this script to make two-way models where both source populations always have 50% ancestry:

source="g/25/mas"|rc3
targ="Udmurt"
target=source[targ,]|anu
source%@%drn(targ)
name=source|rn
source%@%amu
npop=source|nr

i1=2:npop|ax(x:npop)|ul
i2=1:(npop-1)|rep((npop-1):1)

points=(source[i1,]+source[i2,])/2
dist=points|euo(target|t)

ord=dist|o|h5
do=dist[ord]
n1=name[i1][ord]
n2=name[i2][ord]
swaw(n2<n1,n1,n2)

do|rof(3)|pas(n1|spar(n1|nchar|max),n2)|p

A regular R version of the script has about 58% more characters:

source=as.matrix(read.csv("g/25/mas",r=1))
targ="Udmurt"
target=unname(source[targ,])
source=source[!rownames(source)%in%targ,]
name=rownames(source)
source=unname(source)
npop=nrow(source)

ij1=unlist(lapply(2:npop,function(x)x:npop))
ij2=rep(1:(npop-1),(npop-1):1)

points=(source[ij1,]+source[ij2,])/2

dist=sqrt(outer(rowSums(points^2),sum(target^2),"+")-tcrossprod(points,2*t(target)))

ord=head(order(dist),32)
do=dist[ord]
n1=name[ij1][ord]
n2=name[ij2][ord]
n1s=pmin(n1,n2)
n2s=pmax(n1,n2)

writeLines(paste(sprintf("%.3f",do),formatC(n1s,max(nchar(n1s)),,"s"),n2s))

[Nganasankhan]

Here's a simple but slow Awk implementation of the method used by Dienekes. It takes about 2.2 seconds run when the sources consist of all populations in modern averages and the target is a single population.

It requires GNU Awk because it uses `PROCINFO`, and because it uses arrays of arrays which are faster than multidimensional arrays. A version that used multidimensional arrays took about twice as long to run.

$ curl -Lso mas 'https://drive.google.com/uc?export=download&id=1wZr-UOve0KUKo_Qbgeo27m-CQncZWb8y'
$ awk -f twoway.awk <(grep ^Saami, mas) mas
0.012 142 Saami_Kola -42 Russian_Pskov
0.013 139 Saami_Kola -39 Belarusian
0.013 138 Saami_Kola -38 Lithuanian_VA
0.013 138 Saami_Kola -38 Lithuanian_VZ
0.013 138 Saami_Kola -38 Lithuanian_SZ
0.013 143 Saami_Kola -43 Estonian
0.013 138 Saami_Kola -38 Lithuanian_RA
0.013 139 Saami_Kola -39 Lithuanian_PA
0.014 138 Saami_Kola -38 Lithuanian_PZ
0.014 139 Saami_Kola -39 Latvian
0.014 138 Saami_Kola -38 Russian_Smolensk
0.015 141 Saami_Kola -41 Russian_Kaluga
0.015 184 Saami_Kola -84 Russian_Pinega
0.015 172 Saami_Kola -72 Vepsian
0.016 139 Saami_Kola -39 Russian_Voronez
0.016 138 Saami_Kola -38 Ukrainian_Sumy
$ awk -f twoway.awk m=1 n=2 <(grep ^Saami, mas) mas
0.019 78 Saami_Kola 22 Mansi
0.019 80 Saami_Kola 20 Khanty
$ cat twoway.awk
BEGIN{FS=",";if(!n)n=16}
NR==1{targname=$1;for(i=2;i<=NF;i++)target[i-1]=$i}
NR>1{if(/^,/||$1==targname)next;name[++nrow]=$1;for(i=2;i<=NF;i++)source[nrow][i-1]=$i}
END{
  for(i=1;i<=nrow-1;i++)for(j=i+1;j<=nrow;j++){sum=0;for(k=1;k<=NF-1;k++)sum+=(source[i][k]-source[j][k])^2;sourcedist[i][j]=sum^.5}
  for(i=1;i<=nrow;i++){sum=0;for(j=1;j<=NF-1;j++)sum+=(source[i][j]-target[j])^2;targetdist[i]=sum^.5}
  r=0
  for(i=1;i<=nrow-1;i++)for(j=i+1;j<=nrow;j++){
    d12=sourcedist[i][j]
    d1=targetdist[i]
    d2=targetdist[j]
    frac=(d12^2+d2^2-d1^2)/(2*d12^2)
    dist=(d2^2-frac^2*d12^2)^.5
    pop1=name[i]
    pop2=name[j]
    if(frac<.5){frac=1-frac;temp=pop1;pop1=pop2;pop2=temp}
    if(m==1&&frac>1){frac=1;dist=d1<d2?d1:d2} # convert negative weights to 0%
    if(m==3&&frac<=1)continue # reject models that don't have a negative weight
    odist[r]=dist
    ofrac[r]=frac
    opop1[r]=pop1
    opop2[r]=pop2
    r++
  }
  PROCINFO["sorted_in"]="@val_num_asc"
  for(i in odist){printf("%.3f %.0f %s %.0f %s\n",odist[i],100*ofrac[i],opop1[i],100*(1-ofrac[i]),opop2[i]);if(++printed==n)break}
}

From the output above you can see that when I allowed both positive and negative weights, then the best model for Saami was "142% Saami_Kola + 42% Russian_Pskov" at distance 0.012, but when I allowed only non-negative weights, then the best model was "78% Saami_Kola + 22% Mansi" at the higher distance of 0.019. So because the models with the lowest distance had negative weights, then it indicates that Saami are a population which has contributed ancestry to other populations but that Saami themselves are not mixed.

The best models for Nganasans also get negative ancestry, because there are many modern populations which can be modeled as a combination of Nganasans and another population, but Nganasans don't get a good distance fit as a combination of any two modern populations:

$ awk -f twoway.awk <(grep ^Nganassan, mas) mas
0.033 259 Nenets -159 Khanty
0.039 241 Nenets -141 Mansi
0.060 147 Evenk -47 Sakha
0.064 248 Nenets -148 Ket
0.066 142 Yukagir_Tundra -42 Nivkh
0.066 141 Evenk -41 Yakut
0.069 75 Even 25 Nenets
0.071 123 Evenk -23 Daur
0.071 144 Yukagir_Tundra -44 Ulchi
0.071 119 Evenk -19 Xibo
0.072 123 Evenk -23 Hezhen
0.072 79 Even 21 Selkup
0.072 116 Evenk -16 Korean
0.072 131 Evenk -31 Nanai
0.072 116 Evenk -16 Japanese
0.072 114 Evenk -14 Han_Shanghai

Above the model "Nganasan = 259% Nenets - 159% Khanty" has a distance of 0.033, but the model "Nenets = 62% Khanty + 38% Nganasan" has a distance of only 0.013:

$ awk -f twoway.awk <(grep ^Nenets, mas) mas
0.013 62 Khanty 38 Nganassan
0.016 59 Mansi 41 Nganassan
0.021 87 Selkup 13 Nganassan
0.024 90 Selkup 10 Even
0.024 90 Selkup 10 Evenk
0.025 89 Selkup 11 Yukagir_Tundra
0.025 125 Selkup -25 Ket
0.025 62 Ket 38 Nganassan
0.026 88 Selkup 12 Dolgan
0.029 93 Selkup 7 Sakha
0.029 93 Selkup 7 Yakut
0.029 92 Selkup 8 Koryak
0.029 113 Selkup -13 Shor_Khakassia
0.030 111 Selkup -11 Shor_Mountain
0.030 104 Selkup -4 Tajik_Rushan
0.030 104 Selkup -4 Tajik_Ishkashim

Models with negative weights tend to get worse distance than models without negative weights, which I think might be because they are affected multiple times by population-specific drift.

[Nganasankhan]

When I put the default source populations as a target, I put all modern averages as the source, I set the maximum number of models per target to 1, and I allowed both positive and negative weights, I got the following result:

0.007 IND_Roopkund_A = 56% Brahmin_Gujarat + 44% Hakkipikki
0.011 Baltic_EST_BA = 143% Lithuanian_SZ - 43% Austrian
0.013 BRA_Sumidouro_10100BP = 60% Bolivian_Cochabamba + 40% Pima
0.014 CHN_Yellow_River_LN = 107% Han_Shanxi - 7% Shor_Khakassia
0.015 FIN_Levanluhta_IA = 97% Saami + 3% Juang
0.015 RUS_Chalmny-Varre = 65% Saami + 35% Saami_Kola
0.017 IND_Great_Andamanese_100BP = 96% Onge + 4% Cree
0.019 KAZ_Kipchak = 72% Karakalpak + 28% Tajik_Hisor
0.019 Baltic_EST_IA = 61% Latvian + 39% Finnish_East
0.020 Baltic_EST_MA = 145% Estonian - 45% Finnish_Southwest
0.020 ETH_Mota = 192% Ethiopian_Ari_cultivator - 92% Somali
0.021 ZAF_2100BP = 111% Khomani_San - 11% Ogiek
0.023 CZE_Early_Slav = 61% Croatian + 39% Cossack_Kuban
0.023 POL_Unetice_EBA = 86% Swedish + 14% Darginian
0.024 RUS_Nomad_MA = 60% Tatar_Siberian + 40% Tatar_Lipka
0.025 Sargat_IA = 76% Udmurt + 24% Shor_Khakassia
0.026 VK2020_NOR_North_VA_o1 = 68% Saami + 32% Khanty
0.027 RUS_Kusnarenkovo_Karajakupovo_MED = 108% Tatar_Siberian_Zabolotniye - 8% Oroqen
0.028 England_N = 226% Basque_Roncal - 126% Irish
0.028 RUS_Darkveti-Meshoko_En = 161% Georgian_West - 61% Greek_Trabzon
0.030 TKM_Geoksyur_N = 179% Iranian_Mazandarani - 79% Greek_Dodecanese
0.032 RUS_Yakutia_Ymyiakhtakh_LN = 68% Nganassan + 32% Negidal
0.033 RUS_Baikal_N = 83% Khamnegan + 17% Eskimo_Naukan
0.034 RUS_Shamanka_N = 108% Khamnegan - 8% Greek_Trabzon
0.035 TUR_Barcin_N = 479% BelgianC - 379% Icelandic
0.036 CHN_Qihe_11500BP = 55% Hawaiian + 45% Jamatia
0.038 RUS_Tagar = 145% Besermyan - 45% Mari
0.038 RUS_Mezhovskaya = 74% Udmurt + 26% Irish
0.038 CMR_Shum_Laka = 69% Bakola + 31% Hadza
0.038 RUS_Progress_En = 222% Kubachinian - 122% Georgian_Laz
0.038 Corded_Ware_Baltic = 205% Afrikaner - 105% Spanish_Navarra
0.040 TUR_Pinarbasi_HG = 371% French_Auvergne - 271% Orcadian
0.042 GEO_CHG = 332% Georgian_Kart - 232% Greek_Cappadocia
0.042 DEU_Karsdorf_LN = 209% Irish - 109% Basque_Baztan
0.043 RUS_Krasnoyarsk_MLBA_o = 279% Komi - 179% Moksha
0.043 Yamnaya_RUS_Samara = 258% Kaitag - 158% Georgian_Kart
0.049 RUS_Bolshoy_Oleni_Ostrov = 265% Komi - 165% Ukrainian_Lviv
0.050 CHN_Dzungaria_EBA1 = 281% Tabasaran - 181% Georgian_Laz
0.056 DEU_LBK_KD = 213% Italian_Molise - 113% Tabasaran
0.061 RUS_Yana_UP = 50% Finnish_East + 50% Onge
0.061 RUS_Krasnoyarsk_BA = 321% Selkup - 221% Ket
0.064 Levant_Natufian = 220% Yemenite_Al_Bayda - 120% EmiratiB
0.079 RUS_Khvalynsk_En = 333% Kaitag - 233% Georgian_Kart
0.094 VK2020_NOR_North_LN_HG = 199% Ingrian - 99% Albanian
0.098 RUS_Volga-Kama_N = 493% Norwegian - 393% French_Paris
0.099 NOR_N_HG = 730% Russian_Pskov - 630% Ukrainian_Zhytomyr
0.107 RUS_Sosonivoy_HG = 365% Tajik_Ishkashim - 265% Parsi_India
0.108 RUS_MA1 = 253% Tajik_Shugnan - 153% Azerbaijani
0.112 KAZ_Botai = 270% Shor - 170% Buryat
0.113 RUS_Karelia_HG = 1077% Norwegian - 977% English_Cornwall
0.113 RUS_Samara_HG = 765% Ror - 665% Kamboj
0.127 SWE_Motala_HG = 776% Russian_Pskov - 676% Ukrainian_Zhytomyr
0.133 RUS_Tyumen_HG = 371% Tajik_Ishkashim - 271% Parsi_India
0.136 CHN_Tarim_EMBA1 = 310% Tajik_Shugnan - 210% Ezid
0.140 Baltic_LVA_HG = 660% French_Occitanie - 560% Italian_Lombardy
0.152 SRB_Iron_Gates_HG = 817% Spanish_Barcelones - 717% Italian_Lombardy
0.154 RUS_AfontovaGora3 = 330% Shor - 230% Buryat
0.192 LUX_Loschbour = 998% Spanish_Barcelones - 898% Italian_Lombardy

For example this model is pretty unexpected: "0.040 TUR_Pinarbasi_HG = 371% French_Auvergne - 271% Orcadian". And another similar model is "0.035 TUR_Barcin_N = 479% BelgianC - 379% Icelandic".

RUS_Krasnoyarsk_MLBA_o:I6717 has high affinity to Uralic populations, and even here its best model was "0.043 RUS_Krasnoyarsk_MLBA_o = 279% Komi - 179% Moksha".

And also the best model for kra001 didn't get Nganasan or Ymyyakhtakh ancestry but it was "0.061 RUS_Krasnoyarsk_BA = 321% Selkup - 221% Ket". When I put all modern and ancient populations in sources and I allowed both positive and negative weights, then the best model for Selkups was 70% Ket and 30% kra001:

0.019 Selkup = 70% Ket + 30% RUS_Krasnoyarsk_BA
0.020 Selkup = 76% Nenets + 24% Ket
0.020 Selkup = 70% Ket + 30% RUS_Yakutia_Ymyiakhtakh_LN
0.023 Selkup = 70% Ket + 30% Nganassan
0.024 Selkup = 113% Nenets - 13% Nganassan
0.025 Selkup = 89% Nenets + 11% KAZ_Kazakh_steppe_EMBA
0.025 Selkup = 89% Nenets + 11% RUS_Okunevo_BA
0.025 Selkup = 93% Nenets + 7% KAZ_Botai
0.025 Selkup = 90% Nenets + 10% KAZ_Kanai_MBA
0.026 Selkup = 93% Nenets + 7% RUS_Sosonivoy_HG

BTW when I edited my R script so that I replaced `dist=sapply(d2^2-weight^2*d0^2,sqrt)` with `dist=sqrt(d2^2-weight^2*d0^2)`, it reduced the running time of my script from about 130 ms to about 45 ms. So now my script is about a hundred times faster than the original version by Dienekes.

[TonyC]

0.035 Anthony_C_scaled = 66% TUR_Barcin_N + 34% RUS_Progress_En
0.036 Anthony_C_scaled = 64% TUR_Barcin_N + 36% Yamnaya_RUS_Samara
0.042 Anthony_C_scaled = 59% TUR_Barcin_N + 41% DEU_Karsdorf_LN
0.046 Anthony_C_scaled = 50% TUR_Barcin_N + 50% POL_Unetice_EBA
0.046 Anthony_C_scaled = 68% TUR_Barcin_N + 32% CHN_Dzungaria_EBA1
0.046 Anthony_C_scaled = 56% TUR_Barcin_N + 44% Corded_Ware_Baltic
0.047 Anthony_C_scaled = 70% TUR_Barcin_N + 30% RUS_Khvalynsk_En
0.047 Anthony_C_scaled = 54% CZE_Early_Slav + 46% TUR_Barcin_N
0.056 Anthony_C_scaled = 59% TUR_Barcin_N + 41% Baltic_EST_MA
0.057 Anthony_C_scaled = 60% TUR_Barcin_N + 40% Baltic_EST_IA
0.057 Anthony_C_scaled = 73% TUR_Barcin_N + 27% RUS_Samara_HG
0.059 Anthony_C_scaled = 64% TUR_Barcin_N + 36% RUS_Tagar
0.059 Anthony_C_scaled = 73% TUR_Barcin_N + 27% RUS_Volga-Kama_N
0.060 Anthony_C_scaled = 63% TUR_Barcin_N + 37% Baltic_EST_BA
0.060 Anthony_C_scaled = 75% TUR_Barcin_N + 25% RUS_Karelia_HG
0.060 Anthony_C_scaled = 64% TUR_Barcin_N + 36% RUS_Mezhovskaya
0.063 Anthony_C_scaled = 77% TUR_Barcin_N + 23% RUS_Sosonivoy_HG
0.064 Anthony_C_scaled = 75% TUR_Barcin_N + 25% RUS_MA1
0.065 Anthony_C_scaled = 78% TUR_Barcin_N + 22% RUS_AfontovaGora3
0.065 Anthony_C_scaled = 78% TUR_Barcin_N + 22% RUS_Tyumen_HG

[PLogan]

I'm very intrigued by the Denisovans and eagerly look forward to new findings. In the interim, I like to play with their samples.

This was interesting to me...

Denisova_published.DGenisova_published.DG,-0.621007,0.055059,0.009076,0.043691,0.012341,-0.009592,0.167004,-0.119505,-0.022254,0.029618,0.009644,-0.137551,-0.049372,-0.003953,0.005756,0.000627,0.003917,-0.022227,0.015385,-0.004692,-0.002523,0.004277,-0.004141,-0.002682,-0.000819

Compared against Davidski PCA scaled (Ancients).

16,065,946 models:

0.09209 Denisova_published.DG:Denisova_published.DG = 152.8384% CMR_Shum_Laka:I10874_new_all - 52.8384% KEN_Kakapel_3900BP:KPL001
0.09477 Denisova_published.DG:Denisova_published.DG = 154.1014% CMR_Shum_Laka:I10873_new_all - 54.1014% KEN_Kakapel_3900BP:KPL001
0.09479 Denisova_published.DG:Denisova_published.DG = 153.9795% CMR_Shum_Laka:I10873_new_all - 53.9795% TZA_Kisese:I8821
0.09575 Denisova_published.DG:Denisova_published.DG = 148.5851% CMR_Shum_Laka:I10874_new_all - 48.5851% TZA_Kisese:I8821
0.09659 Denisova_published.DG:Denisova_published.DG = 144.7627% CMR_Shum_Laka:I10874_new_all - 44.7627% KEN_Nyarindi_3500BP:NYA002
0.09732 Denisova_published.DG:Denisova_published.DG = 147.9200% CMR_Shum_Laka:I10873_new_all - 47.9200% KEN_Nyarindi_3500BP:NYA002
0.10020 Denisova_published.DG:Denisova_published.DG = 148.3881% CMR_Shum_Laka:I10873_new_all - 48.3881% TZA_Gishimangeda:I13982_new
0.10112 Denisova_published.DG:Denisova_published.DG = 143.9169% CMR_Shum_Laka:I10874_new_all - 43.9169% TZA_Zanzibar_1300BP:I0589
0.10123 Denisova_published.DG:Denisova_published.DG = 148.4585% CMR_Shum_Laka:I10873_new_all - 48.4585% TZA_Zanzibar_1300BP:I0589
0.10155 Denisova_published.DG:Denisova_published.DG = 140.2133% CMR_Shum_Laka:I10873_new_all - 40.2133% KEN_LSA:I8808
0.10162 Denisova_published.DG:Denisova_published.DG = 284.7608% CMR_Shum_Laka:I10873_new_all - 184.7608% CMR_Shum_Laka:I10871_new_all
0.10204 Denisova_published.DG:Denisova_published.DG = 141.4522% CMR_Shum_Laka:I10874_new_all - 41.4522% TZA_Gishimangeda:I13982_new
0.10216 Denisova_published.DG:Denisova_published.DG = 149.6146% CMR_Shum_Laka:I10874_new_all - 49.6146% MWI_Fingira:I4426_new_all
0.10233 Denisova_published.DG:Denisova_published.DG = 135.5768% CMR_Shum_Laka:I10874_new_all - 35.5768% KEN_LSA:I8808
0.10336 Denisova_published.DG:Denisova_published.DG = 242.0726% CMR_Shum_Laka:I10874_new_all - 142.0726% CMR_Shum_Laka:I10871_new_all
0.10356 Denisova_published.DG:Denisova_published.DG = 147.1668% CMR_Shum_Laka:I10873_new_all - 47.1668% BWA_Xaro_1400BP:XAR002
0.10455 Denisova_published.DG:Denisova_published.DG = 143.5018% CMR_Shum_Laka:I10874_new_all - 43.5018% MWI_Hora_9000BP:I2966
0.10481 Denisova_published.DG:Denisova_published.DG = 143.4914% CMR_Shum_Laka:I10874_new_all - 43.4914% MWI_Hora:I19528
0.10551 Denisova_published.DG:Denisova_published.DG = 129.6013% CMR_Shum_Laka:I10873_new_all - 29.6013% ETH_Mota:I5950_new_WGS.SG
0.10571 Denisova_published.DG:Denisova_published.DG = 126.2785% CMR_Shum_Laka:I10874_new_all - 26.2785% ETH_Mota:I5950_new_WGS.SG
0.10573 Denisova_published.DG:Denisova_published.DG = 140.5850% CMR_Shum_Laka:I10874_new_all - 40.5850% MWI_Chencherere:I4421_new_all
0.10651 Denisova_published.DG:Denisova_published.DG = 142.6577% CMR_Shum_Laka:I10874_new_all - 42.6577% MWI_Fingira:I4427_new_all
0.10651 Denisova_published.DG:Denisova_published.DG = 192.5870% CMR_Shum_Laka:I10874_new_all - 92.5870% CMR_Shum_Laka:I10872_new_all
0.10787 Denisova_published.DG:Denisova_published.DG = 204.7692% CMR_Shum_Laka:I10873_new_all - 104.7692% CMR_Shum_Laka:I10872_new_all
0.10809 Denisova_published.DG:Denisova_published.DG = 142.9700% CMR_Shum_Laka:I10873_new_all - 42.9700% MWI_Hora_9000BP:I2966
0.10856 Denisova_published.DG:Denisova_published.DG = 144.0563% CMR_Shum_Laka:I10873_new_all - 44.0563% MWI_Fingira:I4426_new_all
0.10866 Denisova_published.DG:Denisova_published.DG = 133.8179% CMR_Shum_Laka:I10874_new_all - 33.8179% BWA_Xaro_1400BP:XAR002
0.10866 Denisova_published.DG:Denisova_published.DG = 142.3829% CMR_Shum_Laka:I10873_new_all - 42.3829% BWA_Xaro_1400BP:XAR001
0.10995 Denisova_published.DG:Denisova_published.DG = 139.9634% CMR_Shum_Laka:I10873_new_all - 39.9634% MWI_Hora:I19528
0.11015 Denisova_published.DG:Denisova_published.DG = 121.7440% CMR_Shum_Laka:I10874_new_all - 21.7440% ZAF_1200BP:I9134
0.11035 Denisova_published.DG:Denisova_published.DG = 137.9778% CMR_Shum_Laka:I10873_new_all - 37.9778% MWI_Chencherere:I4421_new_all
0.11094 Denisova_published.DG:Denisova_published.DG = 115.8761% CMR_Shum_Laka:I10874_new_all - 15.8761% KEN_Kakapel_900BP:KPL003
0.11130 Denisova_published.DG:Denisova_published.DG = 118.1570% CMR_Shum_Laka:I10873_new_all - 18.1570% KEN_Kakapel_900BP:KPL003
0.11157 Denisova_published.DG:Denisova_published.DG = 138.7723% CMR_Shum_Laka:I10873_new_all - 38.7723% MWI_Fingira:I4427_new_all
0.11186 Denisova_published.DG:Denisova_published.DG = 130.0501% CMR_Shum_Laka:I10874_new_all - 30.0501% BWA_Xaro_1400BP:XAR001
0.11245 Denisova_published.DG:Denisova_published.DG = 125.5008% CMR_Shum_Laka:I10873_new_all - 25.5008% ZAF_400BP:cha001
0.11251 Denisova_published.DG:Denisova_published.DG = 122.4165% CMR_Shum_Laka:I10873_new_all - 22.4165% ZAF_1200BP:I9134
0.11270 Denisova_published.DG:Denisova_published.DG = 121.3061% CMR_Shum_Laka:I10873_new_all - 21.3061% KEN_400BP:I0595
0.11304 Denisova_published.DG:Denisova_published.DG = 114.9980% CMR_Shum_Laka:I10873_new_all - 14.9980% KEN_Pastoral_IA:I12379
0.11325 Denisova_published.DG:Denisova_published.DG = 116.9730% CMR_Shum_Laka:I10874_new_all - 16.9730% KEN_400BP:I0595
0.11333 Denisova_published.DG:Denisova_published.DG = 112.2252% CMR_Shum_Laka:I10874_new_all - 12.2252% KEN_Pastoral_IA:I12379
0.11343 Denisova_published.DG:Denisova_published.DG = 120.1522% CMR_Shum_Laka:I10873_new_all - 20.1522% KEN_IA_Deloraine:I8802
0.11396 Denisova_published.DG:Denisova_published.DG = 120.1187% CMR_Shum_Laka:I10873_new_all - 20.1187% KEN_Kakapel_300BP:KPL002
0.11440 Denisova_published.DG:Denisova_published.DG = 110.1498% CMR_Shum_Laka:I10874_new_all - 10.1498% KEN_MoloCave_1500BP:MOL001
0.11457 Denisova_published.DG:Denisova_published.DG = 117.6727% CMR_Shum_Laka:I10874_new_all - 17.6727% ZAF_400BP:cha001
0.11469 Denisova_published.DG:Denisova_published.DG = 112.3089% CMR_Shum_Laka:I10873_new_all - 12.3089% KEN_MoloCave_1500BP:MOL001
0.11483 Denisova_published.DG:Denisova_published.DG = 110.1363% CMR_Shum_Laka:I10874_new_all - 10.1363% KEN_Pastoral_IA:I12381
0.11484 Denisova_published.DG:Denisova_published.DG = 114.5062% CMR_Shum_Laka:I10874_new_all - 14.5062% KEN_IA_Deloraine:I8802
0.11485 Denisova_published.DG:Denisova_published.DG = 108.3507% CMR_Shum_Laka:I10874_new_all - 8.3507% KEN_MoloCave_1500BP:MOL003
0.11495 Denisova_published.DG:Denisova_published.DG = 117.2378% CMR_Shum_Laka:I10874_new_all - 17.2378% BWA_Taukome_1100BP:TAU001
0.11497 Denisova_published.DG:Denisova_published.DG = 108.0880% CMR_Shum_Laka:I10874_new_all - 8.0880% TZA_PN:I13981

This is compared against the Reich dataset. Had to remove the Ancestor Ref


Target: Denisova_published.DGenisova_published.DG
Distance: 0.5006% / 0.00500627
31.2 Ignore_MbutiPygmy(relative).SDG
15.8 Mbuti.SDG
13.8 Biaka
10.6 Khomani
8.0 South_Africa_2200BP.SG
5.8 Ignore_Mbuti_discovery
5.8 Panama_IsthmoColombian_Colonial_oAfrica_lc.SG
3.0 Papuan.SDG
3.0 South_Africa_1900BP.SG
2.4 South_Africa_2000BP_lc.SG
0.6 Ethiopia_4500BP_published.SG

Distance to:	Denisova_published.DG:Denisova_published.DG
0.03428081	Primate_Chimp:Chimp_HO
0.03486317	Chimp.REF:Chimp.REF
0.03513759	Altai_Neanderthal_published.DG:Altai_published.DG
0.03523505	Altai_Neanderthal.DG:Altai_snpAD.DG
0.03770031	Biaka:HGDP01086
0.03806440	Ignore_Biaka:HGDP00453
0.03837467	Ignore_Biaka.SDG:HGDP00453.SDG
0.04049750	Gorilla:Gorilla
0.04071965	Vindija_Neanderthal.DG:Vindija_snpAD.DG
0.04078252	Biaka.SDG:HGDP01086.SDG
0.04079230	Biaka:HGDP00472
0.04097614	Chagyrskaya_Neanderthal.SG:Chagyrskaya.SG
0.04287544	Biaka:HGDP00475
0.04308039	Biaka:HGDP00460
0.04345890	Biaka:HGDP00469
0.04357520	Biaka.SDG:HGDP00472.SDG
0.04370407	Biaka:HGDP00473
0.04388042	Biaka:HGDP00458
0.04397353	Biaka.SDG:HGDP00460.SDG
0.04406752	Biaka.SDG:HGDP00475.SDG
0.04406810	Biaka.SDG:HGDP00458.SDG
0.04415301	Biaka.SDG:HGDP00469.SDG
0.04439302	Gorilla.REF:Gorilla.REF
0.04510120	Vindija_light:Vindija_light
0.04536374	Biaka:HGDP01094

[PLogan]

Here is a two-way modeling using version 1 (original) of the codebase.

The dataset is the Reich K36 converted to G25 using Nganasan's linear regression a few weeks ago. I removed the Ancestor.REF and other Denisovan samples. Source had 14,307 samples left.

It evaluated over 102 million models.

Denisova_published.DGenisova_published.DG,-0.621007,0.055059,0.009076,0.043691,0.012341,-0.009592,0.167004,-0.119505,-0.022254,0.029618,0.009644,-0.137551,-0.049372,-0.003953,0.005756,0.000627,0.003917,-0.022227,0.015385,-0.004692,-0.002523,0.004277,-0.004141,-0.002682,-0.000819

0.015 -8% Ignore_Dungan_PCA_outlier:UZB-415 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.015 -8% Russia_AngaraRiver_Medieval.SG:irk032.SG + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.015 -9% Ignore_Dungan_PCA_outlier:UZB-415 + 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG
0.015 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG - 9% Mongolia_Dornod_LateMedieval_o:UGO002
0.016 -8% Mongolia_Dornod_LateMedieval_o:UGO002 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG - 9% Russia_AngaraRiver_Medieval.SG:irk032.SG
0.016 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG - 9% Mongolia_Dornod_LateMedieval_o:TSA004
0.016 -8% Mongolia_Dornod_LateMedieval_o:TSA004 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG - 9% Mongolia_Arkhangai_EarlyMedieval_o:OLN001
0.016 -9% China_AmurRiver_N:NE58 + 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG
0.016 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG - 9% Mongolia_Sukhbaatar_XiongnuLateMedieval:TAV006
0.016 -9% China_Xinjian_IA.SG:M8R1.SG + 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG
0.016 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG - 9% Mongolia_Arkhangai_XiongnuEarlyMedieval_3:TUH002
0.016 -11% Ethiopia_4500BP_published.SG:mota.SG + 111% Gorilla:Gorilla
0.016 -8% Mongolia_Arkhangai_LateMedievalEE001 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -8% Mongolia_Uvurkhangai_LateMedieval:NRC001 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -8% Mongolia_Selenge_LateMedieval:BRG001 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -10% Aari.DG:T_Aari-2.DG + 110% Gorilla:Gorilla
0.016 -8% China_Xinjian_IA.SG:M8R1.SG + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -8% Ignore_Xibo.SDG:HGDP01243.SDG + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -7% China_Xinjian_IA.SG:F004.SG + 107% LesCottes_Neanderthal.SG:Les_Cottes_final_provisio nal.SG
0.016 -8% Mongolia_Selenge_XiongnuLateMedieval:BRL001 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -7% Ignore_Ulchi:Ul62 + 107% LesCottes_Neanderthal.SG:Les_Cottes_final_provisio nal.SG
0.016 -8% China_AmurRiver_N:NE58 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -8% Ignore_Xibo:HGDP01243 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -8% Salar:SL6 + 108% VindijaG1_Neanderthal.SG:VindijaG1_final_provision al.SG
0.016 -9% Kyrgyz_China:KZ35 + 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG
0.016 -9% Kyrgyz_China:KZ65 + 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG
0.016 -9% Ignore_Xibo.SDG:HGDP01243.SDG + 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG
0.016 -7% Denmark_Djursland_SingleGraveCulture_lc.SG:RISE128 3.SG + 107% LesCottes_Neanderthal.SG:Les_Cottes_final_provisio nal.SG
0.016 -9% Ignore_Xibo:HGDP01243 + 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG
0.016 -9% Kyrgyz_Kyrgyzstan:Bishkek28443 + 109% Mezmaiskaya2_Neanderthal.SG:Mezmaiskaya2_final_pro visional.SG

[TonyC]

Where are the modern samples?

[PLogan]

Davidski's?

His are always available here... https://eurogenes.blogspot.com/2019/...obal25_12.html

He updates them routinely... think the last was in June.

[Nganasankhan]

The dataset is the Reich K36 converted to G25 using Nganasan's linear regression a few weeks ago. I removed the Ancestor.REF and other Denisovan samples. Source had 14,307 samples left.

You can't get reasonable G25 sims for Neandersovans with K36, because Neandersovans mostly just get the Pygmy component in K36:



Maybe if you did an ADMIXTURE run where the reference samples included chimps and gorillas, then you could project Neandersovans on the run so that it would more accurately reflect the genetic distance between Neandersovans and Sapiens. Or another option would be to revive Neanderthals Jurassic Park style so then we'd get high-quality modern Neanderthal genomes.

Or maybe the most realistic option would be that I finally learned how to use Beagle, so then I could use imputed Neandersovan samples as references for ADMIXTURE.

[rothaer]

The model "100% Finnish = 437% Swedish - 337% Danish" is equivalent to "437% Swedish = 337% Danish + 100% Finnish", or when divided by 4.37 and rounded to the nearest integer, "100% Swedish = 77% Danish + 23% Finnish". (However you cannot actually use this method to find the closest point to Swedes on the line between Finns and Danes, which is actually slightly different.)

Very nice and logical elaborations!