Fig. 5: Taxonomic profiles perform better than functional profiles in discriminating PD from controls. | Nature Communications

Fig. 5: Taxonomic profiles perform better than functional profiles in discriminating PD from controls.

From: Machine learning-based meta-analysis reveals gut microbiome alterations associated with Parkinson’s disease

Fig. 5

Violin plots depict the density of AUC values obtained across ML approaches (within-study cross-validation, CV; study-to-study validation, CSV; and leave-one-study-out validation, LOSO). Within the violin plots the average AUC and the standard deviation (circles and vertical lines) are reported (based on 42 AUCs for CSV and 7 AUCs for CV and LOSO). Performances for ML models (Ridge regression classifiers) built on taxonomic (mOTUs) or various functional profiles are compared as labelled along the x-axis. The horizontal grey dashed line marks the 50% AUC threshold indicating random guessing. KO KEGG orthologous gene families, GMMs gut metabolic modules, GBMs gut-brain axis modules, species taxonomy mOTUs_v3 species profiles.

Back to article page