List of significant metabolites for network analysis

We are running Network Analysis for sets of multivariable data.
Output_39_P1_87_081622.csv (7.6 KB)

Using the attached SAS output as an example. It is from a 2-group study. Mice (male, n=12 per group) were fed a control diet or a high-fat diet (HFD) for 12 weeks. At the end, liver was harvested and analyzed to see metabolomic differences between the 2 groups.

We identified 87 metabolites that met the criteria for statistical analysis. The “Raw_P” column showed 18 metabolites and the FDR column showed 5 metabolites that differed significantly between the two groups.

We understand that we can only use the metabolites that are significantly different between the two groups for network mapping.

Here are our questions.

  1. Should we use FDR (a list of 5) or “Raw_P” (a list of 18) to get a list of significant compounds?
  2. We noticed that the metabolic pathways identified by Global Network Analysis is presented in FDR-adjusted p-values. If we use the list of FDR-adjusted significant compounds (a list of 5) for analysis, our results will be “double FDR-adjusted” and findings seems to be conservative. Is it a correct approach?

Lin Yan, Ph.D.
Grand Forks Human Nutr Res Ctr

The network analysis in a hypothesis generation method. For your questions:

  1. It takes a list of metabolite of interest (for instance, based on similar patterns or fold changes). Raw p values or FDR are just used to select those that are interesting;
  2. This is independent of (unaware of) your upstream tests. The input list is assumed to be the truth.

Overall, your question is more relevant when you combine all the steps under a single statistical modelling and interpret the p values in this context.

Great, thanks for your explanation!

As lomics-type of research turns to be one of our daily routines, MetaboAnalist is a great tool we have enjoyed using it! Many thanks!