Where can I find binaries for previous plink2 versions?

[Dan Gealow]

Older binaries of plink2 used to be available at https://s3.amazonaws.com/plink2-assets/alpha[#]/plink2_[os]_[date].zip -- however, everything older than alpha5 seems to have been removed. Is there anywhere I can currently download these older binaries from? We're moving to the latest "Alpha 5.10 final" for new analyses, but would like to have access to older versions for replicability of previous results.

[Christopher Chang]

The prebuilt <a5.10 binaries have been removed due to the biobank-scale multithreaded --make-bed bug.  However, they are still tagged on GitHub, so you can recompile them from source when necessary for replicability.

[Dan Gealow]

Looking at the release notes: "Data corruption from this bug should be pretty obvious, but if you were in the habit of using --bfile/--make-bed instead of --pfile/--make-pgen with UK Biobank data, you may want to rerun large --make-bed operations and downstream steps to be safe."

What would the corruption look like, how exactly did this bug corrupt data? There's a huge number of data loads and analyses we and our customers have done with plink2 over the past ~5 years, so any way to check which ones are suspect would be incredibly helpful.

Thanks,

Dan

--
You received this message because you are subscribed to a topic in the Google Groups "plink2-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/plink2-users/w4Qe3el6_7w/unsubscribe.
To unsubscribe from this group and all its topics, send an email to plink2-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/plink2-users/1d838331-11c2-48e9-b2af-033f75922a45n%40googlegroups.com.

--

Dan Gealow

Customer Solutions Architect

Paradigm4

The information contained in this message, and any attachments, is confidential and proprietary.  It is intended only for use by the designated recipient. If you have received this communication in error, please notify the sender and destroy the message. Thank you.

[Chris Chang]

You can check sanity of allele frequencies.  Are there blocks of SNPs where they suddenly stop lining up well with expectations?

The bug was a 32-bit integer overflow, where multithreaded —make-bed might rewrite earlier bytes in a 65536-variant block instead of setting later bytes, when those later bytes were past byte 2^32 within the block.

[Chris Chang]

You can focus on —make-bed + <sample filtering or reordering> steps where more than 300k samples remained in the >4 GiB output.  If an AVX2 build was used, only sample reordering was dangerous on UK Biobank-scale data, not sample filtering.