Classification of genomic data: some aspects of feature selection

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Feature selection is crucial for genomic data analysis, enhancing both diagnostic classification and data interpretability.
• Multivariate approaches effectively reduce data dimensionality but can lead to variable subset dependency on classifier objectives.

Purpose of the Study:

• To investigate the impact of different objective functions on feature selection outcomes in genomic datasets.
• To highlight the importance of cautious interpretation of selected features due to potential gene correlations.

Main Methods:

• Applied multivariate feature selection techniques to genomic datasets.
• Evaluated the dependency of selected feature subsets on various classifier objective functions.
• Analyzed gene correlations within selected subsets.

Main Results:

• Demonstrated that the composition of selected gene subsets is contingent upon the specific objective function employed by the classifier.
• Identified that correlated genes can lead to the selection of different subsets for classification tasks.
• Confirmed the necessity of careful interpretation of selected gene sets.

Conclusions:

• The choice of objective function significantly influences feature selection in genomic data.
• Understanding gene correlations is vital for accurate interpretation of classification-related feature subsets.
• Feature selection strategies must consider both predictive power and interpretability in genomic studies.