Subspace learning using structure learning and non-convex regularization: Hybrid technique with mushroom reproduction...

Area of Science:

• Computational Biology
• Machine Learning
• Bioinformatics

Background:

• High-dimensional gene selection is crucial in cancer research.
• Existing feature selection methods (filter, wrapper, embedded) have limitations.
• Hybrid approaches can enhance gene selection performance.

Purpose of the Study:

• To propose a novel hybrid gene selection method combining filter and wrapper strategies.
• To improve the identification of informative genes for cancer datasets.
• To evaluate the proposed method's effectiveness on benchmark datasets.

Main Methods:

• Developed an unsupervised filter-phase method: non-convex regularized non-negative matrix factorization and structure learning (NCNMFSL).
• Utilized mushroom reproduction optimization (MRO) in the wrapper-phase for feature subset selection.
• Combined NCNMFSL for filtering irrelevant features and MRO for selecting discriminative ones.

Main Results:

• The hybrid method achieved high accuracies across multiple cancer datasets (e.g., 0.97 for Breast, 0.98 for Colon and Prostate).
• Achieved top accuracies of 0.97, 0.84, 0.98, 0.95, 0.98, 0.87, and 0.85 on Breast, Heart, Colon, Leukemia, Prostate, Tox-171, and GLI-85 datasets, respectively.
• Demonstrated superior performance compared to state-of-the-art feature selection techniques.

Conclusions:

• The proposed hybrid gene selection method is effective and proficient for high-dimensional cancer datasets.
• The combination of NCNMFSL and MRO offers a powerful approach for identifying key genes.
• This method shows significant potential for advancing cancer research and personalized medicine.