Picrust2 and ANCOM

User Support
1

Hello everyone,
I use QIIME2 quite a lot to do ordinary metagenomics analysis (along the lines of the moving pictures tutorial). This time though I'm dealing with pathway analysis for the first time, I'm trying to grasp the basic concepts but time is short, and documentation is not as complete as for the "standard" analysis.
So, what I think I've got so far (any of these points could be wrong!):

  • I used the Picrust2 qiime2 plugin in "full-pipeline" mode, following the tutorial:
    qiime picrust2 full-pipeline --i-table 'Denoise_DADA2/table.qza' --i-seq 'Denoise_DADA2/rep-seqs.qza' --output-dir Picrust/ --p-threads 30 --p-hsp-method pic --p-max-nsti 2 --verbose

  • I obtained the following files: ec_metagenome.qza, ko_metagenome.qza, pathway_abundance.qza. I turned them into qzv and had a look. All good.

  • Now I need to answer the fundamental question: do any of the KO terms change among my conditions? The first thing I could think of was to do the same thing I usually do for the standard analysis: add pseudocounts to ko_metagenome.qza
    qiime composition add-pseudocount --i-table ko_metagenome.qza --o-composition-table ko_comp-table.qza
    and then perform the ANCOM analysis:
    qiime composition ancom --i-table ko_comp-table.qza --m-metadata-file ../metadata.tsv --m-metadata-column Treatment --o-visualization ancom_KO-Treatment.qzv --verbose

In my head, by doing this I should find out which pathways are significantly differentially expressed between my categories of samples. (spoiler: no results.
Now, this is probably oversimplified if not entirely wrong. So the question is: how can I find out what is going on with pathways and my treatments, starting from the files ec_metagenome.qza, ko_metagenome.qza and pathway_abundance.qza? Is ANCOM the right way to do it?

2

Hello @mtroo,

Welcome to the Qiime 2 forums!
:qiime2: :cat: :airplane:

I appreciate your detailed post and support your approach.
Dada2 -> picrust2 -> differential abundance testing :+1:

I'm not sure if there is a prefered method for testing KOs or pathways, but songbird should provide better results than ANCOM. :bird:
:package: GitHub - biocore/songbird: Vanilla regression methods for microbiome differential abundance analysis
:scroll: Establishing microbial composition measurement standards with reference frames | Nature Communications

Let's see what other people suggest!

Colin

3

Hi Colin, thanks a lot for your reply!
I'll definitely give Songbird a try today.

Also, a follow-up question for you or any other kind readers: let's say that, at the end of this, I know which genera are differentially expressed (standard analysis) and which pathways are differentially expressed (picrust + songbird).

Is there a way to correlate which of those genera are responsible for which pathway (including one-to-many and many-to-one relationships)? Or is it asking too much of this data?

Thanks again!

4

Sure. That sounds like a 'bipartite graph.' Maybe check out this paper.

So you can do it, but I think it is a lot to dry to get out of the data. :crying_cat_face:

So you have amplicons, which show taxonomy and phylogeny of microbes. From this you predict pathways, then try to correlate observed microbes with predicted pathways using a bipartite graph. All from just the amplicons. It's a long shot :dart:

For functional analysis, you really need to transcriptomics, or if you're rich, proteomics :weight_lifting_man:

But, this analysis is worth trying, perhaps as a way to discuss future directions for this research!

Colin