QIIME 2 Meta-Analysis of Rhizosphere Microbiome with Mixed V3-V4/V4 Data: Best Practices for Downstream Analysis

[timanix (Timur Yergaliyev)]

Hi!

@John_Kim_Aligato

• We designed mock primers and trimmed them using cutadapt.
• We confirmed that the only dataset that was labeled as "single-end" read was primarily a paired-end read that was merged. So, we also merged all of the paired-end reads using VSEARCH.
• After that, we consistently truncated the merged reads at position 404 and denoised them using DEBLUR instead of DADA2.
• Then, we merged the feature tables and representative sequences together.
• We classified our representative sequences using "non-v4-16s taxonomy with greengenes2 2024.09 classifier."
• Using our merged feature table, we also filtered the tree from Greengenes 2024.09.phylogeny.id.nwk for us to have a phylogenetic tree that can be used for further downstream analyses.

That sounds good. By following those steps, are you satisfied with the number of reads you retained for the analyses? Also, do the taxonomies assigned make sense to you? I would keep that output if both answers are "Yes".

@John_Kim_Aligato

Our upstream analyses end there, and we will proceed with the downstream analyses such as diversity analyses, differential abundance analysis, co-occurrence network analysis, etc.

If all reads from different studies were trimmed using the same mock primers and merged with VSEARCH (except already merged) and then denoised with Deblur, it should make ASVs directly comparable between the studies that you selected. I would keep that information in the metadata file to check the effect on beta diversity. I would definitely check how PCoA plots look like, and account for study/original primer effects in stat analyses.

@John_Kim_Aligato

Speaking of co-occurrence network analysis, can you recommend the latest or your preferred tools or packages that can make a co-occurrence matrix and visualize them, using our feature table from the upstream analyses as our input?

To be honest, I am not a big fan of network analyses, so my experience with that is not sufficient to recommend the best tool and practices.
But you can check :

• one recommended here
• SCNIC. This one even has the q2-plugin, but it is outdated, so it is better to use the standalone version.
• And here is a guide.

I tried SCNIC a while ago, and it worked for me.

I will leave the topic queued for other mods with better experience with network analyses to add on it.

Best,