What kind of microbiome data can I use? Can I provide transcriptomic data?
Can I calculate metabolic potential scores even if I don’t have metabolomics data?
What about metabolomics features that have not been assigned a compound identification?
How should I normalize my metabolomics data, or will MIMOSA2 normalize it for me?
What does it mean for a taxon to have a negative contribution value?
What cutoff(s) should I use to identify microbial metabolites and contributing taxa?
You can use MIMOSA2 with multiple different types of microbiome data, and MIMOSA2 does not require any specific normalization or quantification. If you can format your data in a way that complies with the input requirements described on the Input Data page, you can use it. Generally, you need either abundances of taxa/populations that can be linked with genomic information, or you need genomic information annotated with KEGG Ortholog abundances. We have not comprehensively assessed whether using amplicon data, metagenomic data, or metatranscriptomic data consistently produces better results.
Technically, you can do this using the get_species_cmp_scores function in the R package. However, we do not recommend it. CMP scores are a very approximate method for
predicting metabolic processes, and they are most useful as a tool for interrogating possible mechanisms that explain observed metabolite trends, rather than
predicting changes in unknown metabolites.
Currently, we can’t do anything about them. You can assign a putative identification and see if MIMOSA2 is able to significantly link them with microbiome data.
MIMOSA2 assumes you have already normalized your metabolite data appropriately for the assay and study design with which it was collected. In particular, MIMOSA2 assumes that you have quantitative measurements for each individual metabolite feature - i.e. differing measurement values or intensities between samples are proportional with underlying differences in metabolite concentrations. MIMOSA2 will optionally log-transform your data, which may improve model fit.
MIMOSA2 and MIMOSA1 answer the same general questions: Which metabolites vary consistently with microbial metabolic potential? What taxa and reactions do they appear to depend on? MIMOSA2 has expanded input and modeling options and uses a more consistent statistical approach for doing so. It also presents analysis results in a more user-friendly format via the web server.
MIMOSA2 assesses whether your observations are consistent with what we know about microbial metabolism from a specific database or set of metabolic reconstructions. Therefore, it can only analyze metabolites that are present in the specific reference database, and that are also found somewhere in the community metabolic model constructed from your microbiome dataset. If you modify the settings for how the community metabolic model is being constructed, you may change whether certain metabolites are included in the model.
The best choice of reference data depends on the type of microbiome data you have and the environment of your samples. Here are various considerations that can affect the answer to this question:
If you have 16S rRNA amplicon data, there is no reason not to try using multiple reference options and assess the effect on your results. Your results may change because the number of taxa that can be mapped to a model and the number of metabolites present in the model can differ widely between the different sources.
Yes. An advantage of running MIMOSA2 via the web server is that you can run analyses using KEGG on our server. The web server only provides a subset of the community metabolic network to download, so you will not be able to run any MIMOSA2 analyses locally unless you have access to files from the KEGG FTP database.
The app has a maximum file upload size of 300 MB. In particular, a stratified microbiome table with many samples can reach this limit relatively easily. If you reach this limit, or if you are able to upload your data but find the web app is slow or unresponsive, we recommend you filter low-abundance and rare taxa from your microbiome dataset to decrease file size. MIMOSA2 automatically filters taxa with both low abundance and low prevalence, as these are very unlikely to be identified as key contributors. Alternatively, you can run your analysis locally using the R package instead of the web app.
“Negatively predicted” metabolites can be interpreted several different ways. Possible reasons for a negative correlation between a metabolite’s levels and its metabolic potential include incorrectly annotated or missing reactions, and effects beyond direct metabolic reactions such as growth promotion or toxicity. In simulations and in a simple validation dataset, we have found that taxonomic contributors to a model with a negative slope are slightly less likely to represent a true taxon-metabolite link than contributors to a model with a positive slope. Contributors identified for negatively correlated metabolites could therefore represent true relationships, but should be interpreted more cautiously than positively correlated metabolites.
The contribution values produced by MIMOSA2 are a measure of each taxon’s importance in explaining variance in measurements of a specific metabolite. The taxa with the largest contribution values, positive or negative, can be considered the largest potential influencers. A positive contribution value indicates that if that taxon were removed, the metabolite would be less variable than is actually observed. A negative contribution value indicates that if that taxon were removed, the metabolite is predicted to be more variable than is actually observed. This means that the negatively-contributing taxon is somehow compensating for or mitigating the predicted metabolic effects of other taxa.
This will depend on analysis goals, study design, and microbiome properties. We typically classify “putative microbe-influenced metabolites” as those predicted with a positive slope and a p-value less than 0.1. The largest taxonomic contributors to variation in those metabolites are considered their “potential taxonomic contributors”, typically any taxon with a contribution to variance greater than 1%.