Mouse 16S model training in MelonnPan article


I am trying to figure out the prediction of metabolites using mouse 16S data via MelonnPan. I looked over the previous posts regarding the 16S analysis with MelonnPan, and sorry if I missed anything regarding my question.

The data submitted in the Github repository is from human gut data. I want to use the mouse gut data to train model, and hence, for the data from Theriot et al, 2014, I sent an e mail to request the metabolomics data that paired with the 16S data uploaded to the NCBI database.

As far as I understand, the KO output of the Picrust2 is not suitable for predict module, which tool or way did you use to train the model for mouse 16S data in MelonnPan article?

Thank you,


Hi @eray-sahin - we used the OTUs from the original publications directly to train the model in the MelonnPan paper. As for the KO output of PICRUSt2, we have not tested it explicitly but you can give it a try. Given that OTUs are highly sparse, there might be an advantage to collapse the OTUs to higher-level informative features (e.g. as described here and here]).

All the best,