Important announcement from Nick Patterson about Admixtools

Angela

Elite member
Messages
21,823
Reaction score
12,325
Points
113
Ethnic group
Italian
Whoever wrote the code, if he comes back after 6 months and read his own code i seriously doubt he will remember what he did, looking at the source code the naming of variables and functions and the organization of the files is horrible though the coding style is clean.
 
See:
Iosif Lazaridis on Twitter: "Nick Patterson asked me to advertise that there is a new version of ADMIXTOOLS on github. https://t.co/7F43hlCzWD This fixes a serious bug in which qpfstats produced erroneous answers if: (cont'd)" / Twitter

Back to the drawing board.

Unbelievable that this would happen.

Was it deliberately introduced, and if so, for what reason.

If so, also these Labs better step up their security and wall off the original algorithms.

What do you expect? These are graduate students that wrote the code, they are not professional programmers. Even professional programmers make mistakes. We do code the way we did code in the 70's. Nothing has changed except the languages. Instead of Fortran and Cobol it's C++ and Javascript.
 
What are the implications of this? Cause to me this does not really change much beyond the aforementioned bug?
So no way this was done deliberately.
That's how coding goes, code, troubleshoot, fix, code ad nauseam.
Admix tools for the most part gave expected results, I wonder what this even changes in the bigger picture.

Disclaimer: Not a coder.

Edit. On top of that ADMIX TOOLS is code from 2012, the bigger surprise is how stable it has been, I know there is Admix Tools 2, and Admix Tools 5.0, but that software has stood the test of time, 9 years longevity in code is underappreciated.
 
What are the implications of this? Cause to me this does not really change much beyond the aforementioned bug?
So no way this was done deliberately.
That's how coding goes, code, troubleshoot, fix, code ad nauseam.
Admix tools for the most part gave expected results, I wonder what this even changes in the bigger picture.

Disclaimer: Not a coder.

Edit. On top of that ADMIX TOOLS is code from 2012, the bigger surprise is how stable it has been, I know there is Admix Tools 2, and Admix Tools 5.0, but that software has stood the test of time, 9 years longevity in code is underappreciated.

Unless we know what the bug is and whether its results are used downstream we won't know the impact on the overall program. Then we have to consider if it was used in studies and the resultant articles.
 

Amazing how some people won't even take Nick Patterson, who is no graduate student, seriously when he specifically says "qpfstats" produced erroneous answers. I don't know how anyone could be more direct.

See, apparently, if the results fit your preconceptions, then the code errors must have been minor and didn't affect the answers at all.

This alert is from this past June. How many people are even aware of it and using the correct version.
 
I’ve noticed samples from older studies don’t match when overlayed with some of 2021 studies and samples (possibly the Daunians too).

I thought some new or updated tool compromised the Standard we rely on … and that some of the 2020 / 2021 “official” PCAs and Results are not fully reliable.
 
So have we seen any article retractions or corrections? Can amateurs re-run the studies given that we have their data?
 
This is in admixtools documentation:

*** there was a bad bug in release 7.0 when qpfstats was run with
1) allsnps: YES
2) inbreed: NO
3) Samples with pseudo-diploid data.

In the new release if you do this, f-statistics involving psuedo-populations twice (such aas an f_3 with target pseudo-diploid are
(correctly) flagged as having very large standard error.


Looks like qpfstats itself was introduced just a year ago with v7.0
 
So have we seen any article retractions or corrections? Can amateurs re-run the studies given that we have their data?

in theory yes, if you know the “official” prep and protocol.

… we don’t have all the data, for example: no Daunians, (their status is unclear to me).
 

This thread has been viewed 3711 times.

Back
Top