Optimized cancer gene selection using armadillo optimization algorithm and support vector machine

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional cancer datasets pose challenges for classification due to irrelevant or redundant features.
• Effective feature selection is critical to enhance accuracy and reduce computational load in cancer data analysis.

Purpose of the Study:

• To develop and evaluate a hybrid AOA-SVM method for efficient and accurate feature selection in cancer datasets.
• To identify minimal, biologically relevant gene markers for improved cancer diagnostics.

Main Methods:

• A hybrid AOA-SVM method was proposed, combining AOA's optimization and diversity maintenance with SVM classification.
• Gene selection involved local optimization within subgroups and a shuffling phase to identify informative gene subsets.
• The method was validated on leukaemia, ovarian, and CNS cancer datasets.

Main Results:

• The AOA-SVM method achieved high accuracy across all tested cancer datasets.
• For the ovarian dataset, 99.12% accuracy and 98.83% AUC-ROC were obtained with 15 genes.
• Perfect classification (100% accuracy) was achieved for leukaemia (34 genes) and CNS (43 genes) datasets.

Conclusions:

• The AOA-SVM hybrid is a highly accurate and computationally efficient tool for cancer diagnostics.
• It demonstrates potential for precision medicine by identifying minimal gene markers.
• The method effectively distinguishes between cancerous and healthy tissues using selected gene subsets.