PDA

View Full Version : FTDNA new practice of predicting R1b subclades



GoldenHind
06-18-2014, 11:34 PM
A few weeks ago I ran across someone in the R1b-P312 project who was now "predicted" to be L238. On checking, this person appears to have done no SNP testing at all, but at least I could why from his markers and matches they made the prediction. I mentioned this on the L238 thread and RMS mentioned he had seen these predictions recently of L21 as well

I have since seen another example where someone previously assigned R1b-M269 status was now "predicted" to be P312. In checking, it seemed likely but far from guaranteed.

Now I have been contacted by someone previously characterized M269 who is now "predicted" to be U152, and has asked for my advice. He appears to have done no SNP testing at all. In checking both STRs and various matches, I can't see any reason for the prediction. Perhaps they match some variety of U152 STR profile I am unfamiliar with.

If this continues, I can see it could become a problem for both FTDNA customers and project administrators. What do others suggest? Should these people order the "predicted" SNP to confirm the prediction, or should they assume the prediction and proceed from there? Should there be a common practice, or should every case be judged on its own merits?

GTC
06-19-2014, 08:42 AM
IIRC there was some discussion from FTDNA (at its most recent Annual Conference?) about changing the way it predicts haplogroups, but I don't recall any details about the process being given.

rivergirl
06-19-2014, 11:45 AM
I have had a new member join the Whalen project who is predicted as R1b L195.
I'm not sure which haplotype this was based on, he does not match the L144/L195 Whalen/Braswell type, seems more like the L195 Weeks type, which are totally different but may be connected, as they are both connected somehow to CTS1751.
If FTDNA are basing this prediction on a match with one person with the L195 SNP, I'm not very confident.

RobertCasey
06-20-2014, 03:46 PM
It is my understanding that FTDNA uses their YSTR matching system for YSNP prediction. They compare all the close matches and see what YSNP testing has been done. They use some kind of percentage of consistent testing results before prediction is made. This is very different from the older YSNP prediction tools based on YSTR values or my L21 prediction tool which is based on 67 marker fingerprint/signatures and binary logistic regression (my tool could be used for non-L21 tested submissions but would have a 10 to 20 % chance of being non-L21). My recommendation for those who want advice on next steps: 1) test the predicted YSNP or some major YSNP clade that it belongs to - L21, P312, etc. for the lowest cost option that have a reasonable chance of being predicted but could easily have a 10 to 20 % error rate (with the expectation that more testing would be required to get to any useful terminal YSNP); 2) recommend the $200 Nat Geo or BritainDNA static test that would be much better. 3) The very old 23andme test that also includes the atDNA test which is still has a lot of good YSNPs but very dated at $100.

You can do you own version of FTDNA YSNP prediction by looking at 67 matches and seeing what has been tested for for close matches. Better yet, use Mike W's spreadsheet which has more YSNP data but is limited to L21 or P312 (or other spreadsheets maintained for other haplogroups).

GTC
06-21-2014, 11:04 AM
Here, from their website, is how they used to make haplogroup predictions:

https://www.familytreedna.com/learn/y-dna-testing/haplogroups-snps/prediction/

Since the recent revision of their Y haplotree I am advised that they now use 37 markers and attempt to predict deeper into the tree. I guess if the kit has only been tested to 12 or 25 markers then they use those.

It seems that the 37 marker predictions are going awry in some (most?) cases.

Probably time to point out to FTDNA that their website needs updating in this regard.

GoldenHind
06-21-2014, 05:50 PM
Here, from their website, is how they used to make haplogroup predictions:

https://www.familytreedna.com/learn/y-dna-testing/haplogroups-snps/prediction/

Since the recent revision of their Y haplotree I am advised that they now use 37 markers and attempt to predict deeper into the tree. I guess if the kit has only been tested to 12 or 25 markers then they use those.

It seems that the 37 marker predictions are going awry in some (most?) cases.

Probably time to point out to FTDNA that their website needs updating in this regard.

Their haplogroup predictions were reasonable enough when FTDNA was issuing predictions such as R1b-M269. However the case I mentioned above went from M269 to predicted U152, and I can't see any basis for it in this particular case. This person has done no SNP testing, and as far as I can tell, none of their very few 67 marker matches has tested U152.

rms2
06-21-2014, 06:44 PM
Most of the predicted L21 have done no SNP testing either, but a number of them have since ordered the L21 test and have gotten positive results. Of course, L21 is super common among men of Isles descent, so maybe it's like shooting fish in a barrel in that case. On the other hand, unless my memory is failing me, a couple of men of Spanish descent recently got L21+ predictions, too, and, upon testing, they also turned out to be L21+ in reality.

GTC
06-22-2014, 01:01 AM
Their haplogroup predictions were reasonable enough when FTDNA was issuing predictions such as R1b-M269. However the case I mentioned above went from M269 to predicted U152, and I can't see any basis for it in this particular case. This person has done no SNP testing, and as far as I can tell, none of their very few 67 marker matches has tested U152.

How does he look for U152 at 37 markers?

GoldenHind
06-22-2014, 06:09 PM
How does he look for U152 at 37 markers?

He has only two matches at either the 37 or 67 marker level, and both are listed with a terminal SNP of M269, as was the person in question until very recently.

GTC
06-23-2014, 12:46 AM
He has only two matches at either the 37 or 67 marker level, and both are listed with a terminal SNP of M269, as was the person in question until very recently.

That's odd. Perhaps time to ask FTDNA to explain how they arrived at U152 in his case.

rms2
06-23-2014, 01:48 PM
That's odd. Perhaps time to ask FTDNA to explain how they arrived at U152 in his case.

That is a good idea, but if I were the individual in question, I would go ahead and spend the $39 and order a U152 test. That's not that much to risk, and if he gets a negative result he can just back up and try P312 and work forward (or back, depending on the outcome) from there.

Or he could go for broke and order the Big Y.

GoldenHind
01-13-2015, 11:44 PM
Just this week I have encountered two different individuals whose R1b-P312 subclade was incorrectly predicted by FTDNA.

The first was predicted to be U152, and started by ordering SNPs directly below U152. When two of those came back negative, he then ordered U152. That came back negative as well.

The next is someone whose Big Y results shows he is DF19. He has a 104/111 same surname match who is predicted to be L21. I think it is highly probable that L21 prediction is inaccurate.

I would assume most of their predictions will prove to be accurate. However these examples underscore the need to confirm them. Unfortunately I suspect a lot of their customers may not even realize their subclade designations are only predictions and not necessarily accurate.

jdean
01-14-2015, 12:53 AM
Just this week I have encountered two different individuals whose R1b-P312 subclade was incorrectly predicted by FTDNA.

The first was predicted to be U152, and started by ordering SNPs directly below U152. When two of those came back negative, he then ordered U152. That came back negative as well.

The next is someone whose Big Y results shows he is DF19. He has a 104/111 same surname match who is predicted to be L21. I think it is highly probable that L21 prediction is inaccurate.

I would assume most of their predictions will prove to be accurate. However these examples underscore the need to confirm them. Unfortunately I suspect a lot of their customers may not even realize their subclade designations are only predictions and not necessarily accurate.

Got a fellow in the Z18 project (confirmed via Geno2) with a close match at 111 who is predicted positive for a subclade of L513, both have 13 at 492 : )

Huntergatherer1066
01-14-2015, 01:07 AM
Got a fellow in the Z18 project (confirmed via Geno2) with a close match at 111 who is predicted positive for a subclade of L513, both have 13 at 492 : )

I think this is also one of my Z18 uncle's very close matches. I was hoping that the person would be repredicted as at least something upstream of Z18 but I guess not.

GTC
01-14-2015, 03:16 AM
I would assume most of their predictions will prove to be accurate. However these examples underscore the need to confirm them. Unfortunately I suspect a lot of their customers may not even realize their subclade designations are only predictions and not necessarily accurate.

It's up to FTDNA, who takes the money, to ensure that the customer clearly understands what is being presented to them.

Customer communication and Information Technology capability are two areas that I expect will never improve at FTDNA until there is change at the very top, as they are cultural weaknesses.

GoldenHind
03-16-2015, 06:02 PM
While searching for possible DF99 candidates I came across a likely group that is predicted to be L20, a subclade of U152. He now has ordered DF99, so we shall see whose prediction is more accurate, FTDNA's or mine. I also found another surname lineage, all of whom appear to share a common ancestry, where some are predicted to be L20 and others L21. My concern continues to be that people will erroneously assume these predictions are accurate and that there is no need to confirm them.

GoldenHind
03-18-2015, 10:04 PM
I received a notice from FTDNA today that they were scrapping their current prediction system, at least for R1b, and returning to their former practice of much broader predictions. I checked the person predicted to be U152>L2>L20 whom i mentioned above, and see he is now merely predicted to be M269. I don't know how many people will receive this notification, but I suspect this may well cause further head scratching among their customers.

MJost
03-18-2015, 10:19 PM
I received a notice from FTDNA today that they were scrapping their current prediction system, at least for R1b, and returning to their former practice of much broader predictions. I checked the person predicted to be U152>L2>L20 whom i mentioned above, and see he is now merely predicted to be M269. I don't know how many people will receive this notification, but I suspect this may well cause further head scratching among their customers.

I read that too and logged on and found, after I ordered my DF13 test on 6/2/12, and it never showed up on my results tree until now, almost three years later. Hooray! Progress.

MJost

rivergirl
03-18-2015, 11:39 PM
I see most of my predicted L144 project members have gone back to M269.
SNP tested L144 is still showing up under U152 instead of L21.
Some Big Y results are not showing up.

Ramdab
05-23-2015, 08:38 PM
I have had a new member join the Whalen project who is predicted as R1b L195.
I'm not sure which haplotype this was based on, he does not match the L144/L195 Whalen/Braswell type, seems more like the L195 Weeks type, which are totally different but may be connected, as they are both connected somehow to CTS1751.
If FTDNA are basing this prediction on a match with one person with the L195 SNP, I'm not very confident.

Hi rivergirl, i am the weeks L195 type (although name now wicks). I have been trying to connect genealogically to the Ireland Whalens group. If you could have him email me [email protected] that would be great. Regards, Dick

Agamemnon
05-24-2015, 09:38 AM
They used to do that with J1 as well, sometimes you'd be assigned to the wrong subclade... Or even "predicted" to be J2!

GoldenHind
05-24-2015, 10:28 AM
While searching for possible DF99 candidates I came across a likely group that is predicted to be L20, a subclade of U152. He now has ordered DF99, so we shall see whose prediction is more accurate, FTDNA's or mine.

Incidentally this person did indeed test DF99+.