Skip to main content
. 2026 Sep 11;14:1927235. doi: 10.3389/fcell.2026.1927235

FIGURE 2.

Two panels summarizing feature selection and model construction. Panel A shows feature selection for 16S rRNA sequencing with a table listing LASSO selecting sixty-six features, Boruta twenty-five, and RFE thirty, and a Venn diagram illustrating the overlap between methods. Panel B shows Olink proteomics sequencing with LASSO selecting ten features, Boruta four, and RFE twenty, and a Venn diagram for overlap. Panel C presents model construction with labeled schematic icons for LR (logistic regression), XGBoost, RF (random forest), SVM (support vector machine), LightGBM, MLP (multi-layer perceptron), KNN (k-nearest neighbors), and DT (decision tree).

Feature selection and model construction. LASSO, Boruta and RFE were utilized to select the core feature for following analysis. (A) The intersection of all three methods yielded a core feature set of 14 variables in the 16S rRNA sequencing dataset. (B) The intersection of all three methods yielded a core feature set of four variables in the Olink proteomics dataset. (C) eight algorithms were utilized in model construction, including LR, XGBoost, RF, SVM, LightGBM, MLP, KNN and DT.