Out of memory for DIA phospho dataset

[rolivella]

Hello, we're trying to run DIANN for a phosphoproteome with 30 raw files with this additional config params:

--var-mods 2
--var-mod UniMod:21,79.966331,STY
--no-norm

At the moment I tested DIANN allocating 64GB but failed.

Do you know how mem should be allocated in order to run this job? Thanks!

[vdemichev]

How does the log look like and how much RAM do you have in the system?

Vadim

[rolivella]

This is the log: rsDIANNsubmission.output.log.zip

I send this job to a cluster queue where I allocated 64GB of mem. After some hours running, the scheduler automatically killed the job: cgroup/OOM: Killed because of memory limit

[vdemichev]

Two comments:

• I suggest to run in silico prediction in a separate step.
• Mass accuracy setting is strange and is likely having a negative effect on performance & RAM consumption
• Are you sure nothing else eats memory there? One of the EOF is after just 15% of the data analysed - this is strange. But the whole analysis can indeed potentially go above 64Gb, yes.

[vdemichev]

Also, can maybe monitor the RAM consumption of the DIA-NN process. DIA-NN itself did not issue a warning, apparently it did not detect potential out of memory event.

[rolivella]

@vdemichev at the end we found a misconfiguration of the mass accuracy that caused the out of memory, as you suggested. I close the issue. Thank you!