Hi everyone! I'm trying to figure out how to process metagenomic data (obtained using qiime2 and picrust2 for qiime) to do some network analysis using igraph or similar (do you have any recommendations?).
Now, unfortunately I can't find a good tutorial about this passage, so do any of you have something which could help me?
I did a co-occurrence network analysis in one of my recent paper, where I constructed a co-occurrence network from metagenomic data. Basically, I used the OTUs table and did pairwise correlations between OTUs (using the Hmisc package) and plotted those correlations as co-occurrence networks with help of igraph.
Here is the paper where I described a bit in detail how I constructed a co-occurrence network.
https://www.mdpi.com/2076-2607/9/9/1943
Milan
Hi Aldo, one thing you have to keep in mind before performing a network analysis from NGS data, is the compositionality of these. For that reason I would recomend you to use some R package that takes this factor into account like Spiec easi, SparCC or Spring. The use co-occurrences statistics like Pearson or Spearman could lead you to wrong assumptions about the correlation between taxa, due to the some characteristics of compositional data (CoDa). I personally use Spiec easi for my networks, but before using it, I would recommend you to clean your dataset from low abundance features, because then your network would be hard to understand.
Regarding the plotting, I personally do it on Cytoscape, but igraph is also a good option.
Here the tutorial of spiec easi. Have fun with it
https://github.com/zdk123/SpiecEasi
Best regards
David Rosado Porto
Hi David! Thank you for your reply, it confirmed I'm going on the right direction. I actually tried to use SpiecEasi, but I'm struggling so much to understand how to give the package the right data. Unfortunately the tutorial on the GitHub page is not enough and not so clear, so if you have some recommendations on where to find a good one let me know, please!
Thank you again
Aldo
Hi Aldo D'Alessandro, attached the main code I used for my networks. You can play around with the nlambda and lambda.min.ratio values. Lower lambdas.min.ratios will produce denser networks, so you will have to try which one works better for you and produce the most stable one. As I told you before I do the plotting with cytoscape not with igraph, so the code shows how to export the network values to plot them with cytoscape.
Hope it helps you.
Best regards
David Rosado Porto
Thank you to everyone for the nice answers!
Thank you Milan Varsadiya , i'm reading your work and it is being pretty useful for the steps I'm missing. Also thanks to it I'm understanding a bit better how to format data.
And thank you David Rosado-Porto , the code you sent me will be useful for sure when I will understand how to get data into SpieacEasi!
Now I'm fighting with the right formatting of csv files for the taxonomy and otu-table, but I'm beginning to see the light at the end of the tunnel :D
If your outputs are from qiime2, you can easily extract the feature table in .biom format and extract the taxonomic information and merge the two objects. From that point you can just create a phyloseq object and use it for the network.
- Badges
- Science topic