Chromosome Painter

[Chris K]

Hello, I was seeking some help related to the Chromosome Painter. Both my parents have totally different ethnic backgrounds i.e. Middle-East / North-Africa vs. North Europe. However, when looking at the Chromosome Painter, it seems to have mixed up both. There are only a few pairs that really show this distinction (i.e. one of the pairs clearly lacking any North African / Middle East ancestry) while the majority seems to have been mixed up. Anyone with the same problem / observation?

[JeffZ]

The ethnicities are estimated from known SNP mutations that can show up on either chromosome in the pair. The pairs need to be phased in order to get consistent results between the them. FTDNA does this by having you link known matches to your tree. This allows them to identify related matches as maternal/paternal/both. The segments containing the reference ethnicity SNPs can then be more consistently associated in the chromosome painter. Note that they do not yet assign which set is maternal or paternal. You would need to do that based on expected ethnicities.

Another thing that can cause mixed results like this is if your parents are closely related. I was working with someone at Ancestry whose parents are first cousins and was getting inconsistent parental match assignments. The common segments were causing his matches to be assigned randomly to each side. That algorithm needs to ignore these segments when it does its phasing. I would expect similar problems with ethnicity phasing but this was before SideView came out.

It is also possibile that your parents share some ethnicities that do not line up with their known pedigrees. Ethnicity estimates are still very much in their early stages as reference groups improve. Past migrations make this difficult.

[Chris K]

Many thanks for your answers JeffZ.
Unfortunately, I do not have any close matches or at least not close enough to be useful for this exercise. Anyhow, this would solve the issue of knowing which of the two belongs to mom or dad but it won’t fix the issue that the ethnicities are mixed. My problem is that my dad’s ethnicity is totally different from my mom’s so I should be able to easily pick their respective part, i.e. one should clearly show my dad’s ethnicity and the other my mom’s. What I see in the Chromosome painter are ethnicities that are clearly from my dad alone but in both parts and this for almost all of the chromosomes (with some exceptions).
Also, they are not closely related (far from) and the origin of their ethnicity is pretty well know (at least 300-400 years on my mom’s side) unless this goes back very far (>1000 years?), but then again, I think I read somewhere that the biggest influence to these results should be your closest relatives so this would be an argument more to justify a clear difference in ethnicity between both my parents.
I was wondering if mislabelling trace percentages could be to blame? But then this should be a problem for everyone i.e. everyone should have some level of unrelated ethnicities in their results.
Still very puzzled....

[JeffZ]

The raw data from the test is a series of unphased SNP (allele) values for each chromosome in a pair. The test is not able to identify which SNP came from which chromosome, just their location within the pair. The phasing has to be done by comparing known sequences of SNPs from the segments of other testers or reference groups. Without this, you will see the mixing between chromosomes in the pairs and within the same chromosome.

Linking close matches helps the most, but even third/fourth cousin matches can be of some use. DNA Painter has a tool that lets you import match and segment information from multiple testing companies and then sort the segments into maternal/paternal. Since you have such distinct ethnicity for each parent, this could help with determining which side each match belongs.

[Chris K]

Great, thanks for this clarification. Let me look into this and see if i can improve the results.

[StefG]

The pairs need to be phased in order to get consistent results between the them. FTDNA does this by having you link known matches to your tree. This allows them to identify related matches as maternal/paternal/both. The segments containing the reference ethnicity SNPs can then be more consistently associated in the chromosome painter.

This is incorrect. According to the MyOrigins 3.0 white paper, FTDNA phases the DNA using the software package "Eagle" in conjunction with their 100K phasing panel (FTDNA internal data, public data such as the 1000 Genomes Project and Human Genome Diversity Project and other public data). The phasing panel is not the same thing as someone's list of matches! It doesn't even use customers' trees or their FF matches, even if some of those matches are linked to the trees! It simply looks at the customer's dataset and segregates the alleles utilizing segments detected between the customer and the reference panel in a 3-step process, initially working with >4cM segments, then with ~1cM windows and finally tidying up the data using 0.3cM windows (Hidden Markov Model HMM). At the end of the day, it's a probabilistic approach. There are always going to be some errors in the phasing and most likely this will explain why some people are observing inconsistencies, like situation which "Chris K" has reported. Eventually long-read sequencing will replace the current methods and the results should be a lot more accurate...

[StefG]

Hello, I was seeking some help related to the Chromosome Painter. Both my parents have totally different ethnic backgrounds i.e. Middle-East / North-Africa vs. North Europe. However, when looking at the Chromosome Painter, it seems to have mixed up both. There are only a few pairs that really show this distinction (i.e. one of the pairs clearly lacking any North African / Middle East ancestry) while the majority seems to have been mixed up. Anyone with the same problem / observation?

Just to let you know, this isn't unusual. Others have observed similar problems with their results and there was some discussion about it when FTDNA brought out their chromosome painter two years ago. I'm basically half South Asian (Indian) via my dad and half western Europe (German) via my mum. Chromosome painter has mixed up some of my DNA too, although maybe not to the same extent as it has for you (in total I have about 12 chromosome pairs with this problem). One of the issues here is genetic variation. Middle-Eastern / North African people are somewhat closer to northern European people from a genetic view than South Asians are (esp. Indians). Even more distinct would be Europeans and East Asians. It's an issue of where populations plot on a PCA and the genetic distance that separates them. The closer they are, the more likely the algorithm will detect tiny segments in common with both populations and hence struggle to phase the DNA correctly.

[JeffZ]

Yes, I should have gone back to read the MyOrigins white paper again instead of assuming they had incorporated a similar phasing process as Ancestry's SideView. I made something of a comment about using reference data in my second post, but thought they were also now making use of the linked matches. Clearly the "Eagle" process has some limitations if it cannot resolve cases like these.

I find it interesting that using very small segments is considered legitimate for phasing the ethnicity estimates but not for matching with other testers. If this phasing is as accurate as claimed, it should also be usable to help reduce false IBD matching.

[Chris K]

Just to let you know, this isn't unusual. Others have observed similar problems with their results and there was some discussion about it when FTDNA brought out their chromosome painter two years ago. I'm basically half South Asian (Indian) via my dad and half western Europe (German) via my mum. Chromosome painter has mixed up some of my DNA too, although maybe not to the same extent as it has for you (in total I have about 12 chromosome pairs with this problem). One of the issues here is genetic variation. Middle-Eastern / North African people are somewhat closer to northern European people from a genetic view than South Asians are (esp. Indians). Even more distinct would be Europeans and East Asians. It's an issue of where populations plot on a PCA and the genetic distance that separates them. The closer they are, the more likely the algorithm will detect tiny segments in common with both populations and hence struggle to phase the DNA correctly.

Many thanks StefG! Now that makes a lot of sense and the fact that he (my dad) is R1b (but Egyptian as well as his ancestors for generations) won't probably make it easier given the close relation between Europeans and R1b in general.