permutations function data not writing properly to .perm file on Mac PLINK1.9

[Matt Neville]

Hi,

I am trying to get the new PLINK 1.9 to run the permutation function for my linear regression analysis running on Mac OSX 10.9.3.  I am using quantitive phenotypes in a separate phenotype file using  --pheno and --all-pheno.  To run the permutation I am running --linear perm.  If I run the phenotypes one at a time using --pheno-name all is fine, however, if i run all the phenotypes using --all-pheno the data is not being written consistently to the .perm file.  For example, I am running the analysis on 97 SNPs from exome chip data.  The first 2 phenotypes from phenotype file have all 97 SNPs written to the file, then it jumps from 97 to 72 of the SNPs written then 53, then 52 then back up to 72, then for many of the phenotypes there is no data only the file headers.  permutations are clearly being run but the data is not being written.  I have run the same files using PLINK 1.07 using the 1.07 --perm function and all works fine so there doesn't seem to be something wrong with my files.  Is there an error in the code or could I be doing something wrong??

regards

matt.

[Christopher Chang]

This sounds like a PLINK 1.9 --all-pheno data management bug; I'll investigate today, and let you know if I need files from you to help me replicate the problem.

[Matt Neville]

Many thanks,  as an additional thought, to speed up my analysis, and because I have a quad core machine I run multiple PLINKs at the same time by adding an & at the end of each line of PLINK commands in a batch file and then running the batch file with $sh. In this case 4 versions or PLINK were running at the same time each going through the same multiple phenotype file with the same list of SNPs but with a different model (e.g. recessive, dominant) or a different gender (e.g. --filter-males).

matt

[Christopher Chang]

The --all-pheno bug should be fixed in the 5 Jun builds.  (I had forgotten to initialize the "number of permutation successes" array to zero.)

When using the multiple-PLINK-instances strategy, you should specify "--threads 1" and an appropriate --memory value; otherwise, the PLINK instances may destructively interfere with each other.

[Matt Neville]

Many thanks,

It seems to work well now and a significant improvement on 1.07.  Also with this speed improvement I probably won't need to run multiple instances of PLINK at the same time so won't need to worry about threads or memory allocation.

matt