PDA

View Full Version : Deep Ancestry using qpAdm



Kurd
05-19-2016, 01:04 PM
This thread is for posting the qpAdm results for our Iranian members, and subsequent discussion. As previously mentioned, fits with chisq values over 2 and tail probabilities less than 80% should not be taken too seriously.

I have purposely kept the left pops and right pops the same for all members, so that the fixed path values (100% modeling) between the various members can be kept comparable. Although some members may not have gotten good fits, their fixed path values are nonetheless informative, and I may attempt a qpAdm based PCA and dendogram, as I believed I have figured out a way to do that. This would be a first as far as I am aware, and it would be interesting to compare the clustering based on it, to the usual ADMIXTURE based PCAs out there (recent ancestry). We may be able to get clustering on a more ancestral level doing so.

A general comment on results obtained using formal methods. I have noticed that some members get uptight (directed at non-Iranian members) when analysis based on formal methods reveals results/ancestry that does not conform to their preconceived ideas/beliefs, or ADMIXTURE, which is based on recent drift. Those members should remember that their written records only go back a few hundred years at best, and that the further back we go in time, the higher the probability that some of your ancestry will deviate from the average ancestry of your ethnic group. This happens for reasons such as voluntary integration of individuals from one ethnic group into another, or forced integration.

I will circle around and re-run the Pashtun members using the same exact base pops used for Kurds and Iranian members under this thread. I will then move on to the other project members. Once I have done almost everyone, I will work on the PCA and dendogram. It will take me a little while to get to everyone, as I do this on a time-permitting, non paid voluntary basis.

Down the road, I hope to start modeling members using some of the newly sequenced genomes, which I incorporated into my dataset when they first came out, but have not had the time to analyze.

From our Iranian members, Jesus was the only one who obtained excellent fits using my base populations. Thus his results are the most realistic.

Kurd
05-19-2016, 01:05 PM
NO
SAMPLE
Andronovo2
Scythian_IA
Chechen
Saudi
Mongola
CHISQ
TAIL PROBABILITY


1
JESUS
0%
0%
30%
70%
0%
0.208
99%


2
JESUS
0%
5%
22%
73%
0%
0.159
98%


3
JESUS
0%
16%
0%
84%
0%
0.464
98%


4
JESUS
0%
0%
30%
70%
0%
0.206
98%


5
JESUS
0%
0%
0%
100%
0%
4.851
43%


6
JESUS
0%
0%
100%
0%
0%
24.936
0%


7
JESUS
100%
0%
0%
0%
0%
30.541
0%


8
JESUS
0%
100%
0%
0%
0%
57.245
0%


9
JESUS
0%
0%
0%
0%
100%
1208.666
0%





0 .Jesus 1
1 Andronovo2 1
2 Scythian_IA 1
3 Chechen 9
4 Saudi 8
5 Mongola 6
6 MbutiPygmy 10
7 Karitiana 12
8 Onge 9
9 Han 33
10 Yoruba 70
11 Ami 10
jackknife block size: 0.05
snps: 529924 indivs: 170
number of blocks for block jackknife: 711
dof (jackknife): 616.942
numsnps used: 26176
codimension 1
f4info:
f4rank: 4 dof: 1 chisq: 0.016 tail: 0.898991679 dofdiff: 3 chisqdiff: -0.016 taildiff:
B:
scale 1 1 1 1
Karitiana 1.002 1.766 0.52 -0.25
Onge 0.706 0.667 -1.36 0.478
Han 1.348 -0.868 -0.548 1.001
Yoruba 0.006 0.133 1.361 1.63
Ami 1.297 -0.816 0.853 -1.025
A:
scale 111.939 1129.126 6322.085 17384.372
Andronovo2 0.459 1.591 -1.17 1.094
Scythian_IA 0.7 1.033 1.753 0.353
Chechen 0.292 0.774 -0.6 -1.659
Saudi -0.134 -0.313 -0.255 0.935
Mongola 2.048 -0.839 -0.365 0.231


full rank 1
f4info:
f4rank: 5 dof: 0 chisq: 0 tail: 1 dofdiff: 1 chisqdiff: 0.016 taildiff:
B:
scale 1 1 1 1 1
Karitiana 1.003 1.77 0.55 -0.308 -0.681
Onge 0.704 0.651 -1.361 0.483 1.412
Han 1.35 -0.861 -0.523 0.966 -1.109
Yoruba 0.005 0.125 1.358 1.633 0.69
Ami 1.295 -0.828 0.854 -1.036 0.914
A:
scale 112.143 1129.243 6301.212 17638.148 69206.007
Andronovo2 0.458 1.592 -1.121 0.983 0.183
Scythian_IA 0.699 1.034 1.823 0.138 -0.32
Chechen 0.289 0.773 -0.54 -1.861 -0.751
Saudi -0.138 -0.313 -0.19 0.74 -2.073
Mongola 2.049 -0.838 -0.307 0.046 0.035


best coefficients: -0.12 0.128 0.33 0.681 -0.019
ssres:
-0.000005107 -0.000076991 -0.000030378 -0.000022618 -0.000040885
-0.119976156 -1.808709321 -0.713659886 -0.531344713 -0.96048877

Jackknife mean: -0.053726673 0.110256633 0.243996791 0.712588779 -0.01311553
std. errors: 0.435 0.377 0.797 0.295 0.062

error covariance (* 1000000)
189406 -10920 -275434 88743 8204
-10920 141952 -149318 36936 -18650
-275434 -149318 634913 -214616 4455
88743 36936 -214616 86732 2205
8204 -18650 4455 2205 3786


fixed pat wt dof chisq tail prob
0 0 1 0.016 0 -0.12 0.128 0.33 0.681 -0.019 infeasible
1 1 2 0.119 0 -0.081 0.033 0.357 0.691 0 infeasible
10 1 2 1.579 0 -0.081 0.033 0.357 0.691 0 infeasible
100 1 2 0.184 0 0.017 0.212 0 0.793 -0.023 infeasible
1000 1 2 0.135 0 -0.114 0 0.472 0.645 -0.003 infeasible
10000 1 2 0.096 0 0 0.123 0.153 0.738 -0.014 infeasible
11 2 3 1.708 0 -1.164 -0.339 2.503 0 0 infeasible
101 2 3 0.322 0.955875 0.079 0.106 0 0.815 0
110 2 3 3.453 0 -6.75 9.208 0 0 -1.458 infeasible
1001 2 3 0.141 0 -0.101 0 0.441 0.659 0 infeasible
1010 2 3 1.579 0 -1.273 0 2.331 0 -0.058 infeasible
1100 2 3 0.59 0.898616 0.171 0 0 0.817 0.012
10001 2 3 0.159 0.983921 0 0.046 0.223 0.73 0
10010 2 3 3.858 0 0 -1.939 2.704 0 0.235 infeasible
10100 2 3 0.188 0 0 0.231 0 0.794 -0.025 infeasible
11000 2 3 0.206 0.976586 0 0 0.298 0.7 0.001
111 3 4 9.131 0 3.011 -2.011 0 0 0 infeasible
1011 3 4 2.405 0 3.011 -2.011 0 0 0 infeasible
1101 3 4 0.743 0.945895 0.204 0 0 0.796 0
1110 3 4 20.71 0 1.129 0 0 0 -0.129 infeasible
10011 3 4 6.576 0 0 -0.718 1.718 0 0 infeasible
10101 3 4 0.464 0.97695 0 0.159 0 0.841 0
10110 3 4 13.526 0 0 1.328 0 0 -0.328 infeasible
11001 3 4 0.208 0.99496 0 0 0.304 0.696 0
11010 3 4 14.438 0 0 0 1.083 0 -0.083 infeasible
11100 3 4 2.347 0.672234 0 0 0 0.963 0.037
1111 4 5 30.541 1.15E-05 1 0 0 0 0
10111 4 5 57.245 4.50E-11 0 1 0 0 0
11011 4 5 24.936 0.000143372 0 0 1 0 0
11101 4 5 4.851 0.434337 0 0 0 1 0
11110 4 5 1208.666 0 0 0 0 0 1
best pat: 0 0 - -
best pat: 1 0 chi(nested): 0.103 p-value for nested model: 0.747916
best pat: 10001 0.983921 chi(nested): 0.04 p-value for nested model: 0.842287
best pat: 11001 0.99496 chi(nested): 0.049 p-value for nested model: 0.825066
best pat: 11101 0.434337 chi(nested): 4.643 p-value for nested model: 0.0311782

Kurd
05-19-2016, 01:08 PM
NO
SAMPLE
Andronovo2
Scythian_IA
Chechen
Saudi
Mongola
CHISQ
TAIL PROBABILITY


1
NK19191
49%
0%
0%
46%
5%
3.949
27%


2
NK19191
67%
0%
0%
34%
0%
5.843
21%


3
NK19191
0%
0%
100%
0%
0%
8.575
13%


4
NK19191
100%
0%
0%
0%
0%
11.106
5%


5
NK19191
0%
100%
0%
0%
0%
34.994
0%


6
NK19191
0%
0%
0%
100%
0%
57.635
0%


7
NK19191
0%
0%
0%
0%
100%
1118.337
0%




0 .NK19191 1
1 Andronovo2 1
2 Scythian_IA 1
3 Chechen 9
4 Saudi 8
5 Mongola 6
6 MbutiPygmy 10
7 Karitiana 12
8 Onge 9
9 Han 33
10 Yoruba 70
11 Ami 10
jackknife block size: 0.05
snps: 502652 indivs: 170
number of blocks for block jackknife: 711
dof (jackknife): 617.461
numsnps used: 26209
codimension 1
f4info:
f4rank: 4 dof: 1 chisq: 0.308 tail: 0.578656933 dofdiff: 3 chisqdiff: -0.308 taildiff:
B:
scale 1 1 1 1
Karitiana 0.96 1.89 0.254 -0.533
Onge 0.622 0.278 -1.797 0.345
Han 1.37 -0.91 -0.04 0.352
Yoruba 0.035 0.419 1.014 1.938
Ami 1.347 -0.588 0.821 -0.846
A:
scale 128.859 1330.23 2541.876 13720.147
Andronovo2 0.274 1.455 -0.271 1.458
Scythian_IA 0.549 1.198 1.111 0.546
Chechen 0.076 0.551 0.714 -1.415
Saudi -0.409 -0.824 1.783 0.684
Mongola 2.11 -0.682 0.024 0.324


full rank 1
f4info:
f4rank: 5 dof: 0 chisq: 0 tail: 1 dofdiff: 1 chisqdiff: 0.308 taildiff:
B:
scale 1 1 1 1 1
Karitiana 0.961 1.911 -0.25 -0.528 -0.284
Onge 0.616 0.246 1.893 0.817 0.557
Han 1.374 -0.861 0.032 0.19 -1.528
Yoruba 0.033 0.415 -0.931 1.99 0.024
Ami 1.345 -0.61 -0.698 -0.241 1.508
A:
scale 129.442 1326.304 2399.175 14751.57 21259.127
Andronovo2 0.251 1.479 0.082 0.04 -1.655
Scythian_IA 0.553 1.185 -1.036 0.986 1.116
Chechen 0.07 0.559 -0.752 -1.984 0.425
Saudi -0.417 -0.808 -1.83 0.265 -0.869
Mongola 2.11 -0.665 -0.075 -0.145 -0.282


best coefficients: 0.674 -0.77 0.598 0.341 0.157
ssres:
-0.000014936 -0.000454832 -0.000178677 -0.000155892 -0.000334797
-0.05450867 -1.659900606 -0.652075749 -0.568925729 -1.221834488

Jackknife mean: 0.337937311 -0.568627554 0.854770091 0.256126601 0.119793552
std. errors: 0.866 0.577 1.385 0.492 0.098

error covariance (* 1000000)
750471 -94709 -1010867 316944 38161
-94709 333040 -256007 61708 -44031
-1010867 -256007 1918491 -634064 -17553
316944 61708 -634064 241637 13774
38161 -44031 -17553 13774 9649


fixed pat wt dof chisq tail prob
0 0 1 0.308 0 0.674 -0.77 0.598 0.341 0.157 infeasible
1 1 2 2.537 0 35.367 9.86 -63.425 19.198 0 infeasible
10 1 2 0.816 0 35.367 9.86 -63.425 19.198 0 infeasible
100 1 2 0.48 0 1.018 -0.725 0 0.539 0.168 infeasible
1000 1 2 1.94 0 2.284 0 -2.757 1.342 0.131 infeasible
10000 1 2 1.178 0 0 -0.88 1.756 -0.021 0.145 infeasible
11 2 3 5.875 0 -3.655 0.72 3.935 0 0 infeasible
101 2 3 5.644 0 0.87 -0.159 0 0.288 0 infeasible
110 2 3 3.106 0 3.569 -2.982 0 0 0.414 infeasible
1001 2 3 3.151 0 8.853 0 -10.956 3.103 0 infeasible
1010 2 3 5.836 0 -1.284 0 2.25 0 0.034 infeasible
1100 2 3 3.949 0.267052 0.488 0 0 0.459 0.052
10001 2 3 6.623 0 0 -0.393 1.592 -0.199 0 infeasible
10010 2 3 1.182 0 0 -0.393 1.592 -0.199 0 infeasible
10100 2 3 9.811 0 0 0.533 0 0.489 -0.022 infeasible
11000 2 3 6.347 0.0959047 0 0 0.692 0.27 0.038
111 3 4 6.682 0 1.691 -0.691 0 0 0 infeasible
1011 3 4 6.288 0 1.691 -0.691 0 0 0 infeasible
1101 3 4 5.843 0.211167 0.665 0 0 0.335 0
1110 3 4 10.738 0 1.024 0 0 0 -0.024 infeasible
10011 3 4 6.902 0 0 -0.181 1.181 0 0 infeasible
10101 3 4 9.906 0.0420486 0 0.471 0 0.529 0
10110 3 4 13.799 0 0 1.225 0 0 -0.225 infeasible
11001 3 4 7.527 0.110533 0 0 0.875 0.125 0
11010 3 4 8.573 0 0 0 1.001 0 -0.001 infeasible
11100 3 4 17.29 0.00169761 0 0 0 0.864 0.136
1111 4 5 11.106 0.0493232 1 0 0 0 0
10111 4 5 34.994 1.51E-06 0 1 0 0 0
11011 4 5 8.575 0.127245 0 0 1 0 0
11101 4 5 57.635 3.74E-11 0 0 0 1 0
11110 4 5 1118.337 0 0 0 0 0 1
best pat: 0 0 - -
best pat: 1 0 chi(nested): 2.229 p-value for nested model: 0.135472
best pat: 1100 0.267052 chi(nested): 1.412 p-value for nested model: 0.234776
best pat: 1101 0.211167 chi(nested): 1.894 p-value for nested model: 0.168699
best pat: 11011 0.127245 chi(nested): 2.732 p-value for nested model: 0.0983427

Kurd
05-19-2016, 01:10 PM
NO

SAMPLE
Andronovo2
Scythian_IA
Chechen
Saudi
Mongola
CHISQ
TAIL PROBABILITY


1
DMXX
59%
0%
0%
41%
0%
6.443
17%


2
DMXX
46%
0%
0%
50%
4%
5.302
15%


3
DMXX
0%
0%
100%
0%
0%
12.911
2%


4
DMXX
100%
0%
0%
0%
0%
13.809
2%


5
DMXX
0%
100%
0%
0%
0%
34.258
0%


6
DMXX
0%
0%
0%
100%
0%
39.706
0%


7
DMXX
0%
0%
0%
0%
100%
1162.819
0%



0 .DMXX 1
1 Andronovo2 1
2 Scythian_IA 1
3 Chechen 9
4 Saudi 8
5 Mongola 6
6 MbutiPygmy 10
7 Karitiana 12
8 Onge 9
9 Han 33
10 Yoruba 70
11 Ami 10
jackknife block size: 0.05
snps: 463351 indivs: 170
number of blocks for block jackknife: 711
dof (jackknife): 613.447
numsnps used: 25478
codimension 1
f4info:
f4rank: 4 dof: 1 chisq: 0.518 tail: 0.471699838 dofdiff: 3 chisqdiff: -0.518 taildiff: 1
B:
scale 1 1 1 1
Karitiana 0.97 1.981 -0.112 -0.209
Onge 0.599 -0.099 2.007 -0.167
Han 1.366 -0.916 -0.054 0.216
Yoruba 0.014 0.232 -0.25 2.118
Ami 1.354 -0.414 -0.945 -0.628
A:
scale 127.602 1296.117 2177.738 11410.428
Andronovo2 0.29 1.452 0.388 0.572
Scythian_IA 0.556 1.273 -0.771 1.632
Chechen 0.086 0.678 -0.703 -1.382
Saudi -0.392 -0.604 -1.937 0.308
Mongola 2.108 -0.67 -0.092 0.075


full rank 1
f4info:
f4rank: 5 dof: 0 chisq: 0 tail: 1 dofdiff: 1 chisqdiff: 0.518 taildiff: 0.471699838
B:
scale 1 1 1 1 1
Karitiana 0.97 1.984 -0.061 -0.264 0.224
Onge 0.598 -0.129 2.064 0.318 -0.516
Han 1.371 -0.865 -0.044 -0.154 1.532
Yoruba 0.013 0.238 -0.242 2.185 0.335
Ami 1.35 -0.493 -0.823 0.184 -1.491
A:
scale 128.458 1286.305 2119.986 10559.291 23102.873
Andronovo2 0.26 1.5 0.124 -0.758 1.446
Scythian_IA 0.559 1.241 -0.751 1.519 -0.527
Chechen 0.081 0.672 -0.74 -1.429 -1.397
Saudi -0.402 -0.58 -1.961 -0.068 0.806
Mongola 2.11 -0.65 -0.161 -0.267 0.168


best coefficients: 1.275 -0.854 -0.314 0.7 0.193
ssres:
-0.000079834 -0.000716403 -0.000302327 -0.000289331 -0.000622969
-0.171551583 -1.539451471 -0.649658594 -0.621732859 -1.338675161

Jackknife mean: 0.169665976 -0.518568345 0.943571298 0.293468524 0.111862546
std. errors: 1.471 0.86 2.217 0.77 0.152

error covariance (* 1000000)
2163680 -353536 -2843925 911317 122465
-353536 738783 -363009 83092 -105330
-2843925 -363009 4914997 -1627276 -80787
911317 83092 -1627276 592265 40602
122465 -105330 -80787 40602 23051


fixed pat wt dof chisq tail prob
0 0 1 0.518 0 1.275 -0.854 -0.314 0.7 0.193 infeasible
1 1 2 2.167 0 6.692 1.266 -10.474 3.516 0 infeasible
10 1 2 1.729 0 6.692 1.266 -10.474 3.516 0 infeasible
100 1 2 0.548 0 1.104 -0.887 0 0.594 0.189 infeasible
1000 1 2 1.508 0 2.752 0 -3.568 1.676 0.14 infeasible
10000 1 2 2.383 0 0 -1.291 2.172 -0.079 0.198 infeasible
11 2 3 5.171 0 -6.546 1.519 6.027 0 0 infeasible
101 2 3 5.704 0 0.962 -0.288 0 0.326 0 infeasible
110 2 3 2.771 0 4.61 -4.192 0 0 0.582 infeasible
1001 2 3 2.488 0 6.014 0 -7.361 2.346 0 infeasible
1010 2 3 5.574 0 -2.573 0 3.506 0 0.067 infeasible
1100 2 3 5.302 0.151 0.456 0 0 0.503 0.041
10001 2 3 8.503 0 0 -0.715 2.072 -0.357 0 infeasible
10010 2 3 2.409 0 0 -1.177 1.988 0 0.189 infeasible
10100 2 3 11.446 0 0 0.501 0 0.525 -0.026 infeasible
11000 2 3 8.719 0.0332751 0 0 0.576 0.385 0.038
111 3 4 6.598 0 1.93 -0.93 0 0 0 infeasible
1011 3 4 6.111 0 -2.165 0 3.165 0 0 infeasible
1101 3 4 6.443 0.168445 0.591 0 0 0.409 0
1110 3 4 12.523 0 1.046 0 0 0 -0.046 infeasible
10011 3 4 8.842 0 0 -0.323 1.323 0 0 infeasible
10101 3 4 11.542 0.0211011 0 0.418 0 0.582 0
10110 3 4 14.186 0 0 1.234 0 0 -0.234 infeasible
11001 3 4 9.901 0.0421319 0 0 0.758 0.242 0
11010 3 4 12.672 0 0 0 1.013 0 -0.013 infeasible
11100 3 4 16.111 0.00287356 0 0 0 0.887 0.113
1111 4 5 13.809 0.0168697 1 0 0 0 0
10111 4 5 34.258 2.12E-06 0 1 0 0 0
11011 4 5 12.911 0.0242316 0 0 1 0 0
11101 4 5 39.706 1.71E-07 0 0 0 1 0
11110 4 5 1162.819 0 0 0 0 0 1
best pat: 0 0 - -
best pat: 1 0 chi(nested): 1.649 p-value for nested model: 0.199061
best pat: 1100 0.151 chi(nested): 3.134 p-value for nested model: 0.0766595
best pat: 1101 0.168445 chi(nested): 1.141 p-value for nested model: 0.285444
best pat: 11011 0.0242316 chi(nested): 6.468 p-value for nested model: 0.0109836

Kurd
05-19-2016, 01:12 PM
I am not sure if the results based on the average of several members (Iranian results) are directly comparable with individual member results, and as such the comparison between Iranian and individual members should be done with a grain of salt.




NO
SAMPLE
ANDRONOVO 503
CHISQ
TAIL PROBABILITY


1
NK19191
100%
11.106
5%


2
DMXX
100%
13.809
2%


3
JESUS
100%
30.541
0%


4
IRANIANS
100%
17.306
0%









NO
SAMPLE
SCYTHIAN IA
CHISQ
TAIL PROBABILITY


1
DMXX
100%
34.258
0%


2
NK19191
100%
34.994
0%


3
JESUS
100%
57.245
0%


4
IRANIANS
100%
51
0%









NO
SAMPLE
CHECHEN
CHISQ
TAIL PROBABILITY


1
NK19191
100%
8.575
13%


2
DMXX
100%
12.911
2%


3
JESUS
100%
24.936
0%


4
IRANIANS
100%
21.16
0%









NO
SAMPLE
SAUDI
CHISQ
TAIL PROBABILITY


1
JESUS
100%
4.851
43%


2
DMXX
100%
39.706
0%


3
NK19191
100%
57.635
0%


4
IRANIANS
100%
136.846
0%









NO
SAMPLE
MONGOLIAN
CHISQ
TAIL PROBABILITY


1
NK19191
100%
1118.337
0%


2
DMXX
100%
1162.819
0%


3
JESUS
100%
1208.666
0%


4
IRANIANS
100%
2501.581
0%

surbakhunWeesste
05-19-2016, 04:43 PM
How can Iranians show Saudi unless those Saudis have some Iranian ancestry? This is very strange!

tippy
05-19-2016, 05:30 PM
Iranians, like all west Asians, tend to show some ancient south west asian affinity/admixture.

Kurd
05-19-2016, 05:39 PM
How can Iranians show Saudi unless those Saudis have some Iranian ancestry? This is very strange!

Iranians and some W Asian have considerable SW Asian admixture. It evens shows up with tools such as ADMIXTURE which are based on recent drift

surbakhunWeesste
05-19-2016, 05:50 PM
Iranians and some W Asian have considerable SW Asian admixture. It evens shows up with tools such as ADMIXTURE which are based on recent drift

So those allele affinity is strictly the SW Asian component like?

Kurd
05-19-2016, 05:53 PM
So those allele affinity is strictly the SW Asian component like?

Yes, and affinity seems to be there even with formal methods such as these that dig deeper in time. After all Saudi ancestors in general are probably a very likely admixing pop for Iranians

DMXX
05-19-2016, 06:29 PM
I don't understand the population selection, here. Particularly the use of Chechens and Saudis.

Why are you choosing admixed modern populations, which we can be abundantly certain didn't contribute to modern Iranians in any significant shape or form, when we have aDNA from West Asia in the form of Kotias and the Anatolian farmers?

I can understand using modern surrogates as Davidski did for the West Asian-related ancestry in Pashtuns through Georgians in his old qpAdm runs to better furnish the match-ups, but this doesn't make much sense at all. It's no wonder the fits don't look particularly robust!

Kurd
05-19-2016, 07:32 PM
I don't understand the population selection, here. Particularly the use of Chechens and Saudis.



I thought I had clearly explained this in my opening post here
I have purposely kept the left pops and right pops the same for all members, so that the fixed path values (100% modeling) between the various members can be kept comparable. Although some members may not have gotten good fits, their fixed path values are nonetheless informative, and I may attempt a qpAdm based PCA and dendogram, as I believed I have figured out a way to do that

I guess it wasn't clear enough. The goal here is not to obtain excellent fits for everyone, as that would take a very long time, certainly more time than I have to spare. Some were lucky and got excellent fits, with others, I spent a few hours trying to improve their fits but was not able to do so. It then struck me that I could take the fixed path results for everyone, and take a jab at creating probably the 1st PCA and dendogram ever based on the fixed paths (100% modeling) for everyone. To do so, the fixed paths for everyone would have to be directly comparable, meaning I would have to use the same set of pops for everyone.

Whereas the fits for some are poor and uninformative, the fixed paths by contrast are informative in that they are a direct comparison of a member to a fixed set of genomes. Thus I can use the fixed path values as dimensions on a 5 dimensional PCA to see where everyone clusters, on a more 'ancestral' basis if you will. I will be able to plot this innovative qpAdm fixed path based PCA and dendogram in a couple of weeks once I have completed the majority of members.

EDIT: After this project I may start something using ancients only (including the Fu et al genomes in my dataset)

surbakhunWeesste
05-19-2016, 08:22 PM
Yes, and affinity seems to be there even with formal methods such as these that dig deeper in time. After all Saudi ancestors in general are probably a very likely admixing pop for Iranians

Cool, so what about Pashtuns and the Indian, north euro and other admixture in them, Kurds don't have that in the same scale but we still get modeled fine with a mordern kurdish sample, help me understand that anomaly please, also regarding the ancestor: which ofc has to do with the haplo groups! If you model kandahari with some selected modern Indian pop from North (west) India or Iran instead of a Kurdish one do you think that will form a good fit?

Kurd
05-19-2016, 08:36 PM
Cool, so what about Pashtuns and the Indian, north euro and other admixture in them, Kurds don't have that in the same scale but we still get modeled fine with a mordern kurdish sample, help me understand that anomaly please, also regarding the ancestor: which ofc has to do with the haplo groups! If you model kandahari with some selected modern Indian pop from North (west) India or Iran instead of a Kurdish one do you think that will form a good fit?

Contrary to what a couple of members have been advocating, I don't believe the fits will be nearly as good, but the proof will be in the result. Pick 2 NW Indian members or pops, and I will do it for you tonight. In other words, I will keep everything the same as your best run, except switch kurd with the 2 samples of your choice, and will post under the pashto thread tonight

surbakhunWeesste
05-19-2016, 09:33 PM
Contrary to what a couple of members have been advocating, I don't believe the fits will be nearly as good, but the proof will be in the result. Pick 2 NW Indian members or pops, and I will do it for you tonight. In other words, I will keep everything the same as your best run, except switch kurd with the 2 samples of your choice, and will post under the pashto thread tonight

Tashakur, my picks are:
Humza
Monkey d luffy
Varun

DMXX
05-19-2016, 11:43 PM
I thought I had clearly explained this in my opening post here

I guess it wasn't clear enough. The goal here is not to obtain excellent fits for everyone, as that would take a very long time, certainly more time than I have to spare. Some were lucky and got excellent fits, with others, I spent a few hours trying to improve their fits but was not able to do so. It then struck me that I could take the fixed path results for everyone, and take a jab at creating probably the 1st PCA and dendogram ever based on the fixed paths (100% modeling) for everyone. To do so, the fixed paths for everyone would have to be directly comparable, meaning I would have to use the same set of pops for everyone.

Whereas the fits for some are poor and uninformative, the fixed paths by contrast are informative in that they are a direct comparison of a member to a fixed set of genomes. Thus I can use the fixed path values as dimensions on a 5 dimensional PCA to see where everyone clusters, on a more 'ancestral' basis if you will. I will be able to plot this innovative qpAdm fixed path based PCA and dendogram in a couple of weeks once I have completed the majority of members.

EDIT: After this project I may start something using ancients only (including the Fu et al genomes in my dataset)

Ah, so "everyone" referred to both Iranians and non-Iranians alike... That appears to be the source of the misunderstanding, as (in combination with the title) it gave the impression that you were modelling Iranians using "ideal" left-groups rather than the generic roster preserved from your earlier experimentation.

khanabadoshi
05-20-2016, 12:32 AM
Tashakur, my picks are:
Humza
Monkey d luffy
Varun

Might I suggest Jam (who is like my father) instead of myself, as he is less mixed? Considering Balq is my grandmother, using me, might skew things? But Jam and I are usually similar, so it may make no difference. Either way, I'll defer to your collective judgements. :)

Kurd
05-20-2016, 12:58 AM
Might I suggest Jam (who is like my father) instead of myself, as he is less mixed? Considering Balq is my grandmother, using me, might skew things? But Jam and I are usually similar, so it may make no difference. Either way, I'll defer to your collective judgements. :)

Ok unless surbakhun indicates otherwise will go with Jam as per your suggestion

surbakhunWeesste
05-20-2016, 01:13 AM
If Kurdwrora has time, I'd ask him to model K. against every possible data. I picked humza because he is a melting pot, monkey because he clearly has some central Asian stuff goin on and Varun because he is a South and an North Indian mix! This should help me understand certain pattern of the algorithmic functions that the program uses. That's all :) thanks.

Ps: he should use jam, balqis and you :P

khanabadoshi
05-20-2016, 01:25 AM
If Kurdwrora has time, I'd ask him to model K. against every possible data. I picked humza because he is a melting pot, monkey because he clearly has some central Asian stuff goin on and Varun because he is a South and an North Indian mix! This should help me understand certain pattern of the algorithmic functions that the program uses. That's all :) thanks.

Ps: he should use jam, balqis and you :P

If the point is the melting pot, then the order goes: me/Hanif > Jam > Balq > Sadia. So better use me then Kurd.

Kurd
05-20-2016, 01:50 AM
If Kurdwrora has time, I'd ask him to model K. against every possible data. I picked humza because he is a melting pot, monkey because he clearly has some central Asian stuff goin on and Varun because he is a South and an North Indian mix! This should help me understand certain pattern of the algorithmic functions that the program uses. That's all :) thanks.

Ps: he should use jam, balqis and you :P

Unfortunately, kurdwrora does not have more than 1 hour to devote to this plus posting the re-run Pashtun member results which will be directly comparable with others. Therefore, a max of 2 samples (1/2 hour).

Other personal business for tonight includes the evening hike in the mountains, and watching part 3 of the HD Pashto drama "za pakhtun yam" which I highly recommend khorjaane and other Pashto speakers watching as it portrays the multiple facets of present day life in Pakhtunkhwa (striving to modernize and deal with terrorism at the same time).

surbakhunWeesste
05-20-2016, 02:57 AM
[
QUOTE=Kurd;158396]Unfortunately, kurdwrora does not have more than 1 hour to devote to this plus posting the re-run Pashtun member results which will be directly comparable with others. Therefore, a max of 2 samples (1/2 hour).

No problem!


Other personal business for tonight includes the evening hike in the mountains, and watching part 3 of the HD Pashto drama "za pakhtun yam" which I highly recommend khorjaane and other Pashto speakers watching as it portrays the multiple facets of present day life in Pakhtunkhwa (striving to modernize and deal with terrorism at the same time).

I watched the first 15 mins of the show... it seems heavily influenced by bollywood? I'd rather watch bollywood itself and they speak heavily urdu influenced pakhto, I LOL'ed at the taliban's acting, Sorry wrora not for me, its too much drama and less substance. Happy Hiking!