Choice of method for heterogenous dataset and network analysis on a small number of ASVs/OTUs

[VivienPichon]

Hi,

I am using SPIEC-EASI through the package NetCoMi and manage to solve most of the technical issues I met thanks to all the great documentation available here. Thank you for all these great didactic efforts! Being relatively new at bioinformatics, I still have some theoretical questions I can't answer by myself on 1) the choice of the best fitted method for my dataset and 2) the validity of a network analysis I want to perform on a very low number of specific ASVs from my dataset.

1) Choice of method My experiment is based on 16S amplicon sequencing, we harvested samples from plant material based on 4 different parameters: 2 Generations, 2 CUltivars, 2 Treatments, 2 COmpartments. In the end I have 5 replicates of each G-CU-T-CO combination, which makes that I have a total of 80 samples, which we expect quite heterogeneous. We are especially interested in comparing the microbiome networks between the treatments, which means that I can get a maximum of 40 samples per network if I pool the samples from the different compartments, generations, and cultivar. In your paper, I saw that the benchmarking showed that the MB method seems much more accurate than glasso, but you also mentioned in another issue that MB could be unstable for heterogeneous samples. I am now wondering if I should avoid using the MB method. Would you have a recommendation on how to choose the best method for specific datasets? Is a method more resistant to false positive than the other?

2) Network analysis on a low number of ASVs In addition to the global microbiome analysis, we are particularly interested in finding co-occurrence between 84 specific ASVs. I'm planning on doing network analysis only on these 84 ASVs, but I realize that it's extremely small compared to the 9000+ ASVs that I got in total. Is a network analysis based on less than 1% of my total ASVs reliable with SPIEC-EASI or would it be heavily biased? Would you have recommendations on how to study a small dataset like this in the best way possible?

Thank you so much for reading me and for your answers 🙏

Best,

Vivien