Merged my FTDNA data with AADR and ran qpAdm: broad NW European + East African/Horn-like signal, but no accepted model yet
I finally got my FTDNA autosomal raw data converted and merged with the AADR 1240K dataset, then ran ADMIXTOOLS 2 / qpAdm models against ancient and modern proxy populations.
Basic workflow:
- Converted FTDNA raw CSV into PLINK PED/MAP, then BED/BIM/FAM.
- Converted AADR v66 1240K EIGENSTRAT/TGENO into PLINK using PLINK2.
- Merged my sample with AADR.
- Fixed the
.fampopulation labels so ADMIXTOOLS could recognize the AADR groups. - Extracted precomputed f2 stats with ADMIXTOOLS 2.
- Tested qpAdm models using different combinations of European, North African, Levantine, Punic, Horn/East African, Nile Valley, Swahili, and Pastoral Neolithic proxies.
The early North African / Levantine / Punic models failed badly. They produced impossible weights like huge positive North African ancestry paired with huge negative Levantine or Roman ancestry, so I treated those as invalid.
The models became more sensible once I added northwest European proxies like English, CEU, and GBR, plus East African / Horn / Nile-related proxies like Kenya_Somali, Sudan_Kulubnarti, Tanzania_Swahili, and Kenya_PastoralN_Nderit.
The best broad signal was consistently:
Northwest European + East African / Nile-Horn / Swahili-Pastoral-like
The most useful feasible proxy models were roughly:
- English + Tanzania_Swahili-oNearEast: ~47.5% English / ~52.5% Tanzania Swahili-oNearEast
- English + Kenya_PastoralN_Nderit: ~41.4% English / ~58.6% Kenya Pastoral Neolithic Nderit
- English + Kenya_Somali: ~57.4% English / ~42.6% Kenya Somali
Important caveat: none of these were formally accepted qpAdm models. The p-values were still extremely low, so I’m treating them as exploratory proxy positions rather than final ancestry proportions.
My takeaway is that with the AADR references I had available, qpAdm could place the broad axis pretty clearly, but it could not find a clean formally accepted source model for a modern mixed individual. Better Ethiopian/Eritrean/Somali/Afar/Oromo/Tigray/Amhara references would probably improve the modeling a lot.