search

cgj-tang / metagenome_metabolome_prediction

Machine learning pipeline for predicting metabolite abundances from metagenomic features
0 stars 0 forks source link

Predicted metagenomes with metabolomes #1

Open Zubair2021 opened 1 year ago

Zubair2021 commented 1 year ago

Hello Tang, Can I use your pipeline to predict metabolome from Picrust2 predicted KO metagenomes for which I do not have any paired data?

cgj-tang commented 1 year ago

Hi Zubair. No you need paired data to train your model, the pipeline doesn't include a prepackaged model. You could try using MelonnPan (which does have a model ready to go), though I don't know how robust the metabolome predictions will be if you're using inferred metagenomes.

Zubair2021 commented 1 year ago

Thank you for your response. I have discussed it with the author of MelonnPann and it uses Uniref90 IDs instead of KOs as input. Do you have an idea of converting KOs to Uniref IDs? The discussion link is as followings: (https://forum.biobakery.org/t/metabolite-prediction-using-16s-microbiome-data/261)

cgj-tang commented 1 year ago

Oh I see. IIRC you can use the regroup_table script in humann2 or 3 to convert gene annotations across different classification schemes. Still strongly recommend that you train your own model if you can