Hi @carli-jones - MelonnPan has two workflows:
MelonnPan-Predict by default uses a pre-trained model that was internally and externally validated in UniRef90 profiles as described in the manuscript. In order to use this default model, the users need to provide functionally profiled metagenomes in the form of UniRef90’s (as described in the tutorial). Currently, we do not support non-UniRef90-based input for the default workflow.
On the other hand, the
MelonnPan-Train workflow allows users to re-train a new model from paired metabolite-microbiome profiles, and the input features can be any microbial sequence features which can be passed to the
MelonnPan-Predict workflow to generate new predictions.
In summary, if you currently do not have paired metabolites with the 16S data, there is not much we can do. But if you do, then you can re-train a new model (different from the default one) and use that model to predict new samples.
Hope that makes sense!