diff options
| author | Pjotr Prins | 2024-09-18 14:43:47 +0200 |
|---|---|---|
| committer | Pjotr Prins | 2024-09-20 10:03:32 +0200 |
| commit | 150c71c59e128e4f1c4d986c178445f85e048c35 (patch) | |
| tree | 3ed4b5cafe7d86d6c0b16fcf205f27016f5a16ad /topics/lmms | |
| parent | 5fa9f32c40f47de7b503843d383f328015435b50 (diff) | |
| download | gn-gemtext-150c71c59e128e4f1c4d986c178445f85e048c35.tar.gz | |
On permutations
Diffstat (limited to 'topics/lmms')
| -rw-r--r-- | topics/lmms/gemma/permutations.gmi | 34 |
1 files changed, 32 insertions, 2 deletions
diff --git a/topics/lmms/gemma/permutations.gmi b/topics/lmms/gemma/permutations.gmi index 1a9cc2e..282d8fb 100644 --- a/topics/lmms/gemma/permutations.gmi +++ b/topics/lmms/gemma/permutations.gmi @@ -427,7 +427,7 @@ So, now we can pass in a trait using JSON. This is probably not a great idea whe Next write the pheno file and pass it in! ``` -./bin/gemma-wrapper --debug --verbose --force --loco --json --lmdb --input K.json -- -g test/data/input/BXD_geno.txt.gz -a test/data/input/BXD_snps.txt -lmm 9 -maf 0.1 -n 2 -debug +./bin/gemma-wrapper --debug --verbose --force --loco --json --lmdb --input K.json -- -g test/data/input/BXD_geno.txt.gz -a test/data/input/BXD_snps.txt -lmm 9 -maf 0.05 -n 2 -debug ``` note the '-n 2' switch to get the second generated column in the phenotype file. We had our first successful run! To run permutations I get: @@ -442,7 +442,37 @@ and, of course, as this reduced file is generated it not available yet. That was ./bin/gemma-wrapper:230:in `block in <main>': Do not use the GEMMA -p switch with gemma-wrapper if you are using JSON phenotypes! ``` -Hmm. This is a bit harder. The call to GWAS takes a kinship matrix and it gets reduced with every permutation. That is probably OK because it runs quickly, but I'll need to remove the -p switch... OK. Done that and permutations are running in a second for 28 BXD! +Hmm. This is a bit harder. The call to GWAS takes a kinship matrix and it gets reduced with every permutation. That is probably OK because it runs quickly, but I'll need to remove the -p switch... OK. Done that and permutations are running in a second for 28 BXD! That implies computing significance in the web service comes into view - especially if we use a cluster on the backend. + +It is interesting to see that 60% of time is spent in the kernel - which means still heavy IO on GEMMA's end - even with the reduced data: + +``` +%Cpu0 : 39.1 us, 51.0 sy +%Cpu1 : 34.0 us, 54.8 sy +%Cpu2 : 35.8 us, 54.5 sy +%Cpu3 : 37.5 us, 49.8 sy +%Cpu4 : 36.0 us, 53.3 sy +%Cpu5 : 29.5 us, 57.9 sy +%Cpu6 : 42.7 us, 44.7 sy +%Cpu7 : 35.9 us, 52.2 sy +%Cpu8 : 27.0 us, 60.7 sy +%Cpu9 : 24.5 us, 63.2 sy +%Cpu10 : 29.8 us, 58.9 sy +%Cpu11 : 25.3 us, 62.7 sy +%Cpu12 : 28.1 us, 58.9 sy +%Cpu13 : 34.2 us, 52.8 sy +%Cpu14 : 34.6 us, 52.2 sy +%Cpu15 : 37.5 us, 51.8 sy +``` + +There is room for more optimization. + +The good news is for a peak we have we find that it is statistically significant: + +``` +["95 percentile (significant) ", 0.0004945423, 3.3] +["67 percentile (suggestive) ", 0.009975183, 2.0] +``` ## Dealing with epoch |