Beagle imputation run time

Question

[joshuaehill]

Hello, I am running beagle5.1 and am wondering about runtime and input limits. I have successfully run a vcf with 24 samples and 1M markers and it took 1m20s. I am now running a vcf of 507 samples and 250k markers. Beagle has been running for 24 hours now. There is no error output generated yet. No out of memory yet. I have it running separately on 2 different machines each is 48 core with either 250G or 500G ram.

I am only running 1 chromosome through at this time.

the command I am running:

java -Xmx200g -jar ~/.conda/envs/beagle/share/beagle-5.1_24Aug19.3e8-0/beagle.jar gt=/data1/users/joshua.hill/19255Swo-total-gvcf-joint/total-gvcf-joint-only-ScWZqcR_8017_12737-mmiss05-maf005.recode.vcf out=/staging/total-gvcf-joint-only-ScWZqcR_8017_12737-mmiss05-maf005-imputation

On the 500G machine I have allocated 450g to it (java -Xmx450g -jar). I can see that all the cores are being used and something is happening but with no output it is hard to tell. From hat I can tell looking at the process, it is taking all the memory being allocated. It is just stuck at the "Window 1 , Study Markers" stage. Not even going through iterations yet.

Thanks josh

[joshuaehill]

Hello,

Adding to my question. It seems on the machine that I allocated 200g of memory has now errored out with the "OutofMemory" error. So now I know that for 507 samples and 250k markers 200g of memory is not enough. I am hoping that the other machine with 450g of memory allocated will work.

Answer 1

[Bioinformatics_NewComer]

Are you doing this per chromosome? Also, are you doing the imputation chromosome-wide? I think it is really hard computationally do run a chromosome wide imputation, rather it is suggested to perform imputation in chunks (3-5-7Mb however works).