💾 Archived View for thebird.nl › blog › 2023 › gemma-precompute.gmi captured on 2023-05-24 at 17:37:52. Gemini links have been rewritten to link to archived content

-=-=-=-=-=-=-

GEMMA precompute

I have been holding off on precompute, because it is a somewhat complex exercise and there was always something else! But now it is on top of my list. See for an earlier writeup:

../2021/07/gemma-on-hpc-part1.gmi

I'll set up a task tracker, but the main steps are:

Phase 1: Display GEMMA values next to QTLReaper (HK)

Pull data from GN database
Transform GEMMA output to efficient file format using lmdb
Run LOCO compute on main server
Adapt GN to find and load lmdb data
Display results

Phase 2: Adapt search & correlations

Adapt so GEMMA values are used

Phase 3: Octopus

Prepare for invoking remote jobs on Octopus

Phase 4: Permutations

Compute permutations on the cluster

Phase 5: Speed up GEMMA

Speed up GEMMA kinship computations & GWA and create new mgamma tool

Phase 6: LOCO and non-LOCO

Display results of both LOCO and non-LOCO in GN

Phase 7: Residuals

With mgamma: pay attention to residuals, phase and effect sizes