Feature Selection and Molecular Classification of Cancer Phenotypes: A Comparative Study

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional gene expression data analysis is essential for developing diagnostic and prognostic tools.
• Feature selection aims to identify the most predictive subset of genes for classification tasks.

Purpose of the Study:

• To compare the predictive capacity of various feature selection methods combined with different classification algorithms.
• To evaluate these combinations on benchmark cancer-related microarray datasets.

Main Methods:

• Comparative study of feature selectors: Chi-Squared, mRMR, Relief-F, and Genetic Algorithms (GA).
• Evaluation with classifiers: Random Forests, PLS-DA, SVM, Regularized Logistic/Multinomial Regression, and kNN.
• Empirical performance assessment on three cancer microarray datasets.

Main Results:

• Data quality relevant to target classes is critical for successful cancer phenotype classification.
• Filter-based feature selection methods demonstrate similar performance to each other.
• Filters achieve comparable or superior results to GA-based wrappers, with greater ease and speed of implementation.

Conclusions:

• Simple, established feature selectors combined with optimized classifiers provide robust performance for gene expression data analysis.
• Complex and computationally intensive methods are not necessarily required for effective cancer classification.