A TRIZ-inspired bat algorithm for gene selection in cancer classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression data is crucial for cancer diagnosis and prognosis.
• Identifying informative genes from large datasets with complex interactions remains challenging.
• Existing metaheuristic-based gene selection methods face difficulties in pinpointing reliable genes.

Purpose of the Study:

• To propose a hybrid filter/wrapper gene selection method named rMRMR-MBA.
• To enhance the identification of informative genes for cancer diagnosis and prognosis.
• To address the challenges posed by high dimensionality and complex gene interactions in gene expression data.

Main Methods:

• A hybrid approach combining a filter (robust Minimum Redundancy Maximum Relevancy - rMRMR) and a wrapper (modified bat algorithm - MBA).
• rMRMR is used to pre-select promising genes.
• MBA acts as a search engine to identify a small, informative gene set.

Main Results:

• The rMRMR-MBA method was evaluated on ten gene expression datasets.
• Performance was assessed by comparing MBA convergence with and without TRIZ optimization.
• Comparative analysis against ten state-of-the-art methods showed superior or competitive results in classification accuracy and reduced gene count.

Conclusions:

• The proposed rMRMR-MBA method demonstrates promising results in gene selection for cancer research.
• It achieves high classification accuracy, contributing significantly to the field.
• The method effectively identifies a small set of informative genes from complex datasets.