Binary hiking optimization for gene selection: Insights from HNSCC RNA-Seq data

• Published in: Expert systems with applications Vol. 268; p. 126404
• Main Authors: Pashaei, Elnaz, Pashaei, Elham, Mirjalili, Seyedali
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 05.04.2025
• Subjects: Adaptive mutation; Classification; Crossover; Gene selection; Hiking optimization algorithm; HNSCC; Hiking optimization algorithm; Gene selection; Crossover; HNSCC; Adaptive mutation; Classification
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2025.126404

•Introduced binary HOA for gene selection in high-dimensional biomedical data.•Enhanced binary HOA with crossover and mutation mechanisms to boost exploration.•Leveraged logit-SPLS model, surpassing SVM and RF in cancer prediction accuracy.•Identified 7-gene subset with 99 % AUC on the GSE6631 dataset for HNSCC.•Identified TGFBR3, HLF, PTK7, SPINK5, CDK1, CKS1B as prognostic biomarkers in HNSCC. As a common challenge, high dimensionality in gene expression data leads to significant computational difficulties in identifying disease markers. To address this issue, this study proposes a novel hybrid gene selection method that incorporates two variants of Hiking Optimization Algorithm (HOA): the Binary HOA (BHOA) and an enhanced version, BHOA-CM. HOA simulates the adaptive behavior of hikers, adjusting their pace based on terrain slope to efficiently reach a summit. BHOA and BHOA-CM employ a hyperbolic tangent transfer function to map continuous values to binary outputs. BHOA-CM also incorporates one-point crossover and self-adaptive mutation operators to enhance the algorithm’s exploitative capabilities in gene selection. The approach begins with Differentially Expressed Genes (DEGs) analysis to identify relevant genes. BHOA and BHOA-CM are then applied with a hybrid classifier that combines Adaptive Sparse Partial Least Squares (SPLS) and Logistic Regression (logit-SPLS) to optimize gene selection performance. Experimental results on six benchmark microarray datasets demonstrate that the proposed method outperforms recent state-of-the-art techniques in classification accuracy while selecting fewer marker genes. Additionally, the method is applied to real head and neck squamous cell carcinoma (HNSCC) RNA-Seq data and online handwriting data from Alzheimer’s disease (AD) patients in the DARWIN dataset, highlighting the algorithm’s adaptability to diverse biomedical challenges. The BHOA and BHOA-CM methods demonstrate significant promise for feature selection in high-dimensional gene expression data analysis, effectively identifying key marker genes for cancer diagnosis and prognosis.