Error with epa-ng/SEPP

[DanielB682]

Hello,
new to picrust2 (biochemistry background). Tutorial worked fine and when running my own samples i ran into the error running epa-ng --tree. I have 32 GB from other comments/issues I saw this is probably not the problem. Also tried using SEPP for placement and ran into the same problem.

The files I am using are from a sequencing project with several samples. we only got one BIOM file and one corresponding, all including fasta file (we got also fasta for each sample but no BIOM files). Wasn't sure if Picrust2 would even be able to create different files for each samples but just ran it to try. I assume the files are just too big, right? And would there be a relative easy way for a novice to separate this so I can reduce the size of the file and run each sample separately? Or maybe some other approach/tipps?

Thanks already in advance.

[R-Wright-1]

Hi there,

A couple of things to try:

• What was the output that you got, and from which command exactly? Could you try re-running the same command with the --verbose flag added to confirm that is the issue?
• Which version of PICRUSt2 are you using?
• Could you send the files that you're using and I can confirm whether they work?

Thanks,
Robyn

[DanielB682]

This is with verbose flag added.
I use 2.5.3, gonna update i guess ^^
For the files should i attach them to the comment or?
The commands i used:
-picrust2_pipeline.py -s cleanfeature.fasta -i featureTable.sample.total.absolute.txt.biom -o picrust2_out_pipeline -p 1
-picrust2_pipeline.py -s cleanfeature.fasta -i featureTable.sample.total.absolute.txt.biom -o picrust2_out_pipeline -p 1 --placement_tool sepp
-place_seqs.py -s ../cleanfeature.fasta -o out.tre -p 1 \ --intermediate intermediate/place_seqs

and this is the message i get

[SoniramDK]

Hello there, sorry for jumping right in the middle of the conversation but I get the exact same error.

I execute:

it runs for 4 to 6 hours and then returns this error:

Error running this command:
epa-ng --tree /mnt/ssd2/KaradedosDM/new_picrust2/picrust2-2.4.1/picrust2/default_files/prokaryotic/pro_ref/pro_ref.tre --ref-msa picrust2_out_all_barcodes/intermediate/place_seqs/ref_seqs_hmmalign.fasta --query picrust2_out_all_barcodes/intermediate/place_seqs/study_seqs_hmmalign.fasta --chunk-size 5000 -T 1 -m /mnt/ssd2/KaradedosDM/new_picrust2/picrust2-2.4.1/picrust2/default_files/prokaryotic/pro_ref/pro_ref.model -w picrust2_out_all_barcodes/intermediate/place_seqs/epa_out --filter-acc-lwr 0.99 --filter-max 100

The version of picrust2 that I am using is the latest (v2.6.0)

[R-Wright-1]

OK so @DanielB682: yes, you could attach the files to your comment, or send them to me robyn.wright@dal.ca if you prefer. You could also try running only the place_seqs.py command and changing the --chunk_size parameter - the default for this is 5000, so you could try 1000, e.g.:

place_seqs.py -s cleanfeature.fasta -o placed_seqs.tre -p 1 \
              --intermediate placement_working --chunk_size 1000

@SoniramDK: you don't mention how much RAM you have, but 173,341 is a LOT of sequences. Have you carried out any filtering prior to trying to run PICRUSt2? It is common with OTU clustering methods to have an inflated number compared with ASV methods (I compared these in a project I did a while back and found that on the same data, I'd get ~6,000 ASVs or ~70,000 OTUs), so especially with OTU methods I'd recommend some prevalence filtering prior to running PICRUSt2. I'd also consider using ASVs rather than OTUs, unless you have no other option.

[SoniramDK]

Hello again and thanks for the reply.

The available RAM in my computer is 38gb, which apparently is not enough.

Filtering could be a solution indeed. Do you have any suggestions about how to perform the filtering? Also, perhaps you know how much RAM I need for the 173.341 OTUs?

ASVs are not an option in my case.

Thanks again!

[R-Wright-1]

What filtering is most appropriate will depend on your study, study design, number of samples, average read depth, etc, so I can't give a one-size-fits-all answer, but some things to consider:

• Removing samples with low read numbers (popular cut-offs are >2500, 5000 or 10,000 sequences per sample)
• Removing low prevalence OTUs - e.g. removing OTUs that aren't present in at least 10% of samples
• Removing low abundance OTUs - e.g. removing OTUs with less than a maximum count within samples of 10 or 100, or 0.1% of the median read depth

For my own analyses, I often use a combination of these and investigate the impact that each of these have, with the aim of achieving a balance of removing low quality samples/sequences while not removing so many samples that my analyses are impossible to carry out.

[R-Wright-1]

I will just also add that if you'd like to determine whether PICRUSt2 works fine or not, you can try running with the tutorial dataset, as I mentioned in the Google group here, e.g.

wget http://kronos.pharmacology.dal.ca/public_files/picrust/picrust2_tutorial_files/chemerin_16S.zip
unzip chemerin_16S.zip
cd chemerin_16S
place_seqs.py -s seqs.fna -o out.tre -p 1 --intermediate intermediate/place_seqs

If this runs, then it definitely suggests that you have an issue with memory for your dataset.

[DanielB682]

place_seqs.py -s cleanfeature.fasta -o placed_seqs.tre -p 1 \
              --intermediate placement_working --chunk_size 1000

Hey sorry for the late answer, was a busy week. This worked out, thank you :) . So i guess i just follow up with the other steps from the tutorial?

[R-Wright-1]

Oh great!! I'm not sure which version of PICRUSt2 you ended up using - if v2.5.3 then you can follow the workflow here, and if you used v2.6.0 then you can follow the workflow here.

Robyn

[DanielB682]

continued 2.5.3 because i am really inept with terminal usage, and couldn't figure out how to easy update the version. But the thanks for the help and the tutorial. Should I close this or do you want to keep it open for SoniramDK?

[R-Wright-1]

Haha fair enough. No problem!

I'll close it - they had also asked a similar question on the picrust-users Google group, and as I think their issue is related to memory rather than being an actual issue with PICRUSt2, that's probably the more appropriate place for it anyway.