Fig. 2: Dimension reduction techniques applied to the Paroni Sterbini dataset. | Nature Communications

Fig. 2: Dimension reduction techniques applied to the Paroni Sterbini dataset.

From: Nonlinear machine learning pattern recognition and bacteria-metabolite multilayer network analysis of perturbed gastric microbiome

Fig. 2: Dimension reduction techniques applied to the Paroni Sterbini dataset.

The plots represent the best dimension reduction results based on PSI-PR projection-based separability index (PSI) for the three different labels (P-treated, untreated H+ and untreated H−), evaluated in the 2D embedding space. Moreover, also the average values of all pairwise PSI-ROC are reported as overall estimators of separation between the groups in the 2D reduced space. a PCA; b MDS with Bray-Curtis dissimilarity (MDSbc); c MDS with weighted UniFrac distance (MDSwUF); d nonmetric MDS with Sammon Mapping (NMDS); e MCE. Blue dots represent PPI-treated samples, while red and green dots are the untreated samples which resulted either negative (red) or positive (green) to the H. pylori test (histological observation and urease test). f The curves in three different colours (red, blue and green) highlight the different distributions of the three groups on the second dimension for the MCE plot (e).

Back to article page