Why Beagle v5.4 fails during phasing large genotypic data set?

Question

[Famf]

I am trying to phase large genotypic data set (~330 samples and ~25 millions SNPs) by using Beagle v5.4. My command is this:

java -Xmx160g -jar ./beagle.22Jul22.46e.jar gt=./lines_ch1.vcf.gz ref=./ref_lines.bref3 chrom=1 map=./genmap_ch1.map nthreads=60 window=10.0 out=./pased_lines_ch1 impute=false

The job ends when it reachs the window 10 with the following error:

Window 10 [1:52402376-72298110]

Reference markers: 1,624,703
Study markers: 1,624,703

java.lang.OutOfMemoryError: Java heap space at phase.HmmParamData.<init>(HmmParamData.java:73) at phase.PhaseLS.lambda$getParamEst$1(PhaseLS.java:129) at phase.PhaseLS$$Lambda$588/0x00007f767127f568.run(Unknown Source) at java.base/java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:515) at java.base/java.util.concurrent.FutureTask.run(FutureTask.java:264) at java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1128) at java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:628) at java.base/java.lang.Thread.run(Thread.java:834)

Terminating program.

I have been decreasing the window size from the default value 40 to 10, but it still fail. Is it my data too big for being handled by Beagle? Or am I using the wrong phasing parameters? Thanks

[bk11]

This is the memory issue. You may have to specify a temp directory to fix it.

java -Djava.io.tmpdir=/home/temp/ -Xmx160g -jar ./beagle.22Jul22.46e.jar ......

Answer 1

[4galaxy77]

I've not seen this error before, but it is memory/thread related. I think you're running out of memory because it's trying to phase a very large number of markers in a single window. A million markers per window is a lot, even though you don't have a lot of samples. I'd suggest reducing the window size right down to 5cM and maybe increase the Java heap size. -Xms

Otherwise, email Brian Browning, the author, as this might be a specific error related to threading or something - he is pretty responsive.