Discriminative biomarker selection using hybrid multi-population evolutionary computation

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Deoxyribonucleic acid (DNA) sequencing advancements necessitate improved methods for analyzing high-dimensional microarray data.
• Conventional gene selection techniques face challenges in identifying optimal biomarkers efficiently for disease identification.
• Accurate cancer classification and biomarker discovery are crucial for effective disease management.

Purpose of the Study:

• To propose a novel hybrid method, MPKGSA, for robust cancer classification and biomarker discovery.
• To address the limitations of conventional gene selection in handling high-dimensional, low-sample-size microarray data.
• To enhance the efficiency and accuracy of identifying minimal, biologically relevant gene biomarker subsets.

Main Methods:

• Utilized Kernel Principal Component Analysis (KPCA) for initial data dimensionality reduction, preserving biological patterns.
• Developed a Multi-Population Gravitational Search Algorithm (MPKGSA) incorporating Opposition-Based Learning (OBL).
• Implemented OBL within GSA to enhance search space exploration and prevent premature convergence for diverse solution generation.

Main Results:

• MPKGSA demonstrated a superior balance between convergence and diversity in search.
• Achieved high prediction accuracy using minimal biomarker subsets across six cancer microarray datasets and one breast cancer SNP dataset.
• Outperformed existing meta-heuristic methods in selecting a small, biologically relevant set of gene biomarkers.

Conclusions:

• The MPKGSA method is effective for precise cancer identification and classification.
• The identified gene biomarkers are strongly correlated with biological response classes.
• This approach offers a significant advancement in analyzing complex genomic data for disease research.