Improving feature selection performance for classification of gene expression data using Harris Hawks optimizer with...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression profiling is vital for cancer identification and classification.
• Selecting discriminative feature genes from complex gene expression data is challenging.
• Existing methods often struggle with efficient and accurate feature gene selection.

Purpose of the Study:

• To propose a novel metaheuristic approach, the variable neighborhood learning Harris Hawks optimizer (VNLHHO), for effective gene feature extraction.
• To enhance the selection of highly discriminative feature genes for tumor classification.
• To improve the accuracy and efficiency of identifying cancer-related genes.

Main Methods:

• Utilized F-score for initial gene selection to narrow down the feature set.
• Developed a variable neighborhood learning strategy to balance exploration and exploitation in optimization.
• Incorporated mutation operations to prevent local optima and enhance population diversity.
• Employed a novel activation function for binary conversion and a naive Bayesian classifier as a fitness function.

Main Results:

• VNLHHO achieved classification accuracy exceeding 96.128% for colon, nervous system, and lung tumors.
• Achieved 100% classification accuracy for other tested tumor types.
• Demonstrated superior performance compared to seven other algorithms in classification accuracy, fitness value, and AUC value.

Conclusions:

• The VNLHHO is a highly effective method for feature gene selection in cancer classification.
• This approach significantly improves the accuracy of identifying tumor-specific genes from expression data.
• VNLHHO offers a superior alternative to existing algorithms for gene expression-based cancer research.