A new feature selection method based on feature distinguishing ability and network influence.

Area of Science:

• Bioinformatics
• Systems Biology
• Genomics

Background:

• Disease development involves biomolecule dysfunction and altered interactions.
• Identifying key molecules from omics data is crucial for disease diagnosis and drug discovery.

Purpose of the Study:

• To propose a novel feature selection algorithm, FS-DANI, for identifying important biomolecules.
• To discriminate between different disease conditions using molecular and network-level data.

Main Methods:

• FS-DANI evaluates individual distinguishing ability based on feature effective range overlap.
• Network influence is assessed using module importance and feature centrality.
• A sequential forward search (SFS) determines the crucial feature subset based on comprehensive weights.

Main Results:

• FS-DANI outperformed six other feature selection methods in accuracy, sensitivity, and specificity across ten omics datasets.
• Ablation experiments confirmed the algorithm's effectiveness.
• FS-DANI identified two key microRNAs (miRNAs) for gastric cancer detection with high AUC values.

Conclusions:

• Integrating molecular and network-level evaluations enhances the identification of high-performance disease biomarkers.
• FS-DANI offers a robust method for biomarker discovery in omics data analysis.