diff options
| author | Pjotr Prins <pjotr.public01@thebird.nl> | 2026-11-18 12:18:09 +0000 |
|---|---|---|
| committer | Pjotr Prins <pjotr.public01@thebird.nl> | 2026-11-18 12:18:09 +0000 |
| commit | c1a4e3270e70f91e4e9cd6c7b67a669fbe2f77aa (patch) | |
| tree | 5199ff6f15df93b36aa1f10fc3db894a3f8b31a8 /LICENSE | |
| parent | f026a7b808fd2d61e64d4d53c540fd66fffcff81 (diff) | |
| download | pangemma-master.tar.gz | |
The lmdb path wrote '-' in the assoc's allele1/allele0 columns, so a consumer
that reads them -- genenetwork4 returns GEMMA's columns verbatim -- could not
tell an lmdb scan from a BIMBAM one by the payload. The columns are now taken
from the store's marker record, which gemma-wrapper's geno2mdb.py writes as
"name<TAB>allele1<TAB>allele0" (BIMBAM's second and third field). Stores built
before that hold the name alone, and those two columns still read '-', so an
existing store keeps working unchanged.
Verified on the BXD fixture through LOCO: pangemma on a Gf store and GEMMA on
the same BIMBAM agree on the header, on 636 rows, and on every column --
allele1/allele0 included ('X'/'Y' in both), no differing cell at all. With a
Gb store the same comparison differs on 62 beta / 48 se / 32 l_mle / 76 p_lrt
of those 636 rows, which is Gb's 0.5-as-63/127 quantisation and not this
change: BXD is a mean-genotype panel, and Gb is exact only for integer calls
(0/1/2 -> 0/127/254). Choose Gf for mean genotypes.
Diffstat (limited to 'LICENSE')
0 files changed, 0 insertions, 0 deletions
