Composition-aware correlation of ASVs from 2 different amplicons

General Discussion
,
1

Hi,

I am opening this post (I was missing being the one who asks!) to expose an idea I have had for a project we are finishing in our laboratory. I think it's likely that I'm wrong in some part of my reasoning and that my approach doesn't make sense, so I thought it would be best to share it with expert eyes here.

Briefly, our project involved soil sampling and then sequencing two libraries: one for 16S and one for fungi (ITS2). I finished the analysis for each amplicon, but we are now curious whether there is any correlation between certain bacterial groups (biological control agents) and certain groups of fungi (mycotoxin producers).

I recently came across the SECOM¹ ² methodology developed by the ANCOM and ANCOM-BC authors, and I found it ideal for my analysis.

The thing is, a previous step for aplying this method would be to combine my 16S and ITS2 feature tables. I imagine this can be easily done within or outside QIIME 2, but I wonder if it makes sense to combine amplicons here. Combining ASVs from different marker genes would result in a complete mess if you don't understand what you are doing, but in this case I only want to see correlation based on abundance, so the actual sequence of the ASVs wouldn't be needed.

I think my reasoning is fine, but maybe I'm missing something obvious regarding combining amplicons that would potentially make my results worthless. Any thoughts on this?

Many thanks in advance!

Sergio

--

¹ I put the post under General Discussion because SECOM is related to ANCOM which is already part of QIIME 2, but feel free to move the post under Other Bioinformatics Tools if you feel I'm being too lenient with categories.

² Recently I found that it was suggested to add SECOM to q2-composition. Is this going to happen? I also saw that there is still discussion about the best way of adding ANCOMBC2 to QIIME 2, so I suppose that before adding new stuff like SECOM you may want to prioritize setting a standard for abundance methods.

3 Likes
2

Hello Sergio,

Do you have any data in addition to amplicons?

My former colleague Hyun Seob-Song did this sort of interaction modeling with metabolomics. While I was not part of his team, my memory was that he needed metabolite data because it provided evidence for and against interactions and a falsifiable mechanism that he could not get from correlations alone.

Here's what that looks like:

2 Likes
3

Hi @salias ,
I have not used it myself, but it strikes me that the cross-domain interaction functionality in Speac-Easi may be exactly what you are looking for:

2 Likes
4

Hi @colinbrislawn and @Nicholas_Bokulich ,

No @colinbrislawn , unfortunately amplicon is all I have! I agree with the idea that amplicon itself may be not enough to state that some causal correlation exist (shotgun data or metabolite data would be a great addition), but for now I think I'm happy doing some sort of overview possible correlation analysis. All that correlation stuff would be a side result, the main point of the paper is going to be the amplicon analysis and some phylogenetic inference I already performed.

@Nicholas_Bokulich thank you for the reference! I'll check the method in detail to see if I can use it for my project. The idea is to check if some condition (conventional VS organic farming) affects the BCA bacteria - toxigenic fungi interaction (if any). So I think that filtering my feature tables by that condition and performing two of this interaction methods would do the trick :grimacing:

Sergio

1 Like