blob: 45aa6c93d2c41ba5e9271fda82f9fd130ca21496 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
|
# Databases Getting Out of Wack
This issue refers to precomputed scores generated by the ancient reaper module that runs as a script:
=> https://github.com/genenetwork/genenetwork2/blob/testing/scripts/maintenance/QTL_Reaper_v6.py
We'll create a new issue:
=> lmm-precomputed-scores.gmi
## Tags
* assigned: pjotrp
* priority: high
* type: bug, enhancement
* status: unclear
* keywords: database, gemma, reaper
## Let's use Gemma instead of Reaper
Zachary:
> If we're using GEMMA, we'll need to recalculate all other trait Max LRS scores using
> GEMMA as well (so I think we should just do this with qtlreaper for now). Otherwise
> we'll just have a bunch of qtlreaper scores mixed with GEMMA scores without the user
> having any way of knowing the difference. Also, storing the full results (what Rob
> calls the "full vector model") will require some sort of fundamental change to the
> way we store this data and should be postponed for later (since the high priority
> immediate issue is to ensure that the stored Max LRS values aren't wrong)
> As for Mean, that should be simple, since it's just taking the average of sample values
> immediately after an update.
> The main thing I'm uncertain how to do (though I know is possible since Bonface already
> did something like this with GEMMA) is making the code run in the background after an
> update. It's probably more simple than I'm thinking, though.
Pjotr:
> Current qtlreaper runs in one of Arthur's scripts globally.
> When we recompute on uploading the data we can use GEMMA. That is the
> plan. But let the team do qtlreaper first. Don't want too many moving parts.
Rob:
> Yes, that and more. These values are displayed in search results and used
> to sort by expression mean and peak LRS (using pathetically old code and
> genotypes). Now if GEMMA were wicked fast we could recompute the 60 million
> BXD vector results and store that as a big juicy TRANSFORMATIVE blob of
> data. A big paper in doing just that. Reaper is just wrong at this
> point. We have LMM: We should use it.
|
