Since melonnpan was using human 16s to build datasets, is it possible to use melonnpan to predict the metabolites from rat’s fecal? Because our budget is limited, we can’t afford to do metabolites analysis by NMR spectrum or LC/MS/MS. All we had are 16s datasets from NGS. I’m new in this field, if I’ve did some mistakes please tell me. Thank you!
Hi @kencat - without an environment-specific trained model and paired metabolites data, it’s a long shot to use MelonnPan’s default prediction model which (as you alluded to) was trained on human samples. Consider using a functional inference tool like PICRUSt which seems to be the best option for your setting!
Thanks for your quick replication. Since my NGS supplier had already done th PICRUSTs analysis from my fecal sample, I’m wondering is it possible to use some prediction model to predict metabolites instead of analyzed KEGG pathway and figure out possible metabolites had changed during my experiments. And now I understand it is not suitable for my data. Can I ask another question? If I had 16s NGS data from human fecal sample , can I run MelonnPan’s default prediction model which already pre-trained and predicted possible metabolites or other model I can try? Thank you.
Hi @kencat - ideally MelonnPan is more suitable for metagenomics functional data given that we validated the model using HUMAnN-generated UniRef90 profiles. The only way to use MelonnPan for 16S data is to re-train a new MelonnPan model (assuming you have paired metabolites) to generate predictions in new samples.
Thanks you! If I have human fecal 16S NGS data, I’ll try MelonnPan to run the predictions. Thanks a lot.