sigFeature: Novel Significant Feature Selection Method for Classification of Gene Expression Data Using Support...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Biological data generation is accelerating, necessitating advanced interpretation methods.
• Accurate classification of biological data, especially differentially expressed genes, is crucial for understanding treatment responses.
• Existing feature selection algorithms like Support Vector Machine-Recursive Feature Elimination (SVM-RFE) may yield biologically insignificant results.

Purpose of the Study:

• To introduce a novel feature selection algorithm, sigFeature, that combines Support Vector Machine (SVM) and t-statistic.
• To enhance the discovery of differentially significant features for binary classification.
• To evaluate the performance of sigFeature against existing algorithms using public microarray datasets.

Main Methods:

• Development of the sigFeature R package with a core 'sigFeature' function for automatic feature selection.
• Application of sigFeature to six publicly available Gene Expression Omnibus (GEO) microarray datasets.
• Comparative analysis of sigFeature against three other feature selection algorithms.
• Gene set enrichment analysis for downstream evaluation of biological significance.

Main Results:

• sigFeature demonstrated superior classification accuracy with a smaller set of selected features compared to other algorithms.
• The algorithm successfully predicted the biological signature of four out of six tested microarray datasets.
• Gene set enrichment analysis confirmed the biological relevance of features selected by sigFeature.

Conclusions:

• sigFeature offers a robust and accurate approach for feature selection in biological data analysis.
• The algorithm outperforms traditional methods like SVM-RFE in identifying biologically significant features.
• sigFeature provides a valuable tool for interpreting complex biological datasets and understanding treatment effects.