ECFS-DEA: an ensemble classifier-based feature selection for differential expression analysis on expression profiles.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Differential expression analysis is crucial for identifying sample-distinguishing features.
• Existing methods like multiple hypothesis testing may miss complex explanatory features.
• Multivariate methods face computational challenges, and random forest may not suit all sample distributions.

Purpose of the Study:

• To develop an ensemble classifier-based feature selection tool for differential expression analysis.
• To address limitations of existing methods, particularly concerning varying sample distributions.
• To provide an intuitive and effective tool for identifying important variables in expression profiles.

Main Methods:

• Developed ECFS-DEA, an ensemble classifier for feature selection in differential expression analysis.
• Incorporated a graphical user interface for selecting diverse base classifiers to accommodate sample distribution differences.
• Proposed a common variable importance measure inspired by random forest, applicable to any base classifier.
• Utilized interactive feature selection, k-means clustering for projection heatmaps, and ROC curves for validation.

Main Results:

• ECFS-DEA effectively performs feature selection for differential expression analysis on expression profiles.
• The tool demonstrates applicability across different sample distributions.
• Interactive visualization tools, including projection heatmaps and ROC curves, intuitively show the effectiveness of selected features.

Conclusions:

• Ensemble classifiers offer a robust approach to feature selection in differential expression analysis.
• ECFS-DEA proves effective for analyzing expression profiles with diverse sample distributions.
• The ECFS-DEA software is publicly available for researchers.