A deep-learning framework combined with archetypal analysis to identify functional archetypes in the adult human gut microbiome. Applied to 9,838 whole-genome metagenomic samples from 29 countries.
Code and analyses for the publication — deep archetypal analysis of functional gut microbiome profiles. Includes pre-trained model for archetype assignment.
Bioinformatics pipeline to process whole-genome metagenomic profiles into relative pathway abundances using Kraken2/Bracken, HUMAnN3, and ComBat-seq batch correction.
Key features
→
Compendium of 9,838 WGM samples from 29 countries — processed via Kraken2/Bracken, HUMAnN3, and ComBat-seq batch correction
→
Deep learning embeddings of functional profiles using scAAnet framework
→
Archetypal analysis identifies 3 functional boundary states
→
Archetypes correlate with microbiome stability and confound disease signatures in T2D, CRC, and IBD