Running out of memory from performing isoform discovery and quantification under low-memory mode

[chingyinwan]

Hello. I am trying to use Bambu v3.8.0 to perform isoform discovery and quantification on 8 BAM files with sizes ranging from 30 GB to 52 GB. The server I am running Bambu on has 256 GB of memory. However, the Bambu process ran out of memory even under its low-memory mode. The log messages I got before it crashed are as follows:

--- Start generating read class files ---
< warning messages not included here >
--- Start extending annotations ---
Using a novel discovery rate (NDR) of: 0.47
--- Start isoform quantification ---

Running journalctl -rx, I see an Out of memory: Killed process... log message.

I am calling bambu() with all 8 BAMs passed into it and with quant = TRUE, discovery = TRUE and lowMemory = TRUE. The BAMs only contain primary alignments and contain no secondary or supplementary alignments. What can I do to optimize Bambu's memory usage?

Thank you!

[abcdtree]

I have similar problem. I have bam files with 380G and assigned 800G RAM for the process. I found the step makes the program to crash is actually the output step.

I am trying to output novel transcript only to see whether it can help.

@chingyinwan Please let me whether you have already found some solution.

Thanks

[ch99l]

Hi,

The OOM issue is likely caused by running bambu on eight BAM files simultaneously. To optimise memory usage, I suggest running bambu serially on each BAM file to generate the intermediate read class file (rcFile). You can refer to the "Storing and using preprocessed files (rcFiles)" section of the README for more information. After converting each BAM file to a rcFile, you can then run bambu on the 8 rcFiles simultaneously with quant = True and discovery = True.

Hope this helps!