Gene selection using hybrid binary black hole algorithm and modified binary particle swarm optimization

• Published in: Genomics (San Diego, Calif.) Vol. 111; no. 4; pp. 669 - 686
• Main Authors: Pashaei, Elnaz, Pashaei, Elham, Aydin, Nizamettin
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.07.2019
• Subjects: Algorithms; Binary black hole algorithm; Binary particle swarm optimization; data collection; discriminant analysis; Gene expression; Gene Expression Profiling - methods; Gene Expression Regulation, Neoplastic; Gene selection; genes; Humans; least squares; microarray technology; neoplasms; Neoplasms - classification; Neoplasms - genetics; Sparse partial least squares discriminant analysis; Binary particle swarm optimization; Gene selection; Gene expression; Binary black hole algorithm; Sparse partial least squares discriminant analysis
• ISSN: 0888-7543; 1089-8646; 1089-8646
• DOI: 10.1016/j.ygeno.2018.04.004

In cancer classification, gene selection is an important data preprocessing technique, but it is a difficult task due to the large search space. Accordingly, the objective of this study is to develop a hybrid meta-heuristic Binary Black Hole Algorithm (BBHA) and Binary Particle Swarm Optimization (BPSO) (4-2) model that emphasizes gene selection. In this model, the BBHA is embedded in the BPSO (4-2) algorithm to make the BPSO (4-2) more effective and to facilitate the exploration and exploitation of the BPSO (4-2) algorithm to further improve the performance. This model has been associated with Random Forest Recursive Feature Elimination (RF-RFE) pre-filtering technique. The classifiers which are evaluated in the proposed framework are Sparse Partial Least Squares Discriminant Analysis (SPLSDA); k-nearest neighbor and Naive Bayes. The performance of the proposed method was evaluated on two benchmark and three clinical microarrays. The experimental results and statistical analysis confirm the better performance of the BPSO (4-2)-BBHA compared with the BBHA, the BPSO (4-2) and several state-of-the-art methods in terms of avoiding local minima, convergence rate, accuracy and number of selected genes. The results also show that the BPSO (4-2)-BBHA model can successfully identify known biologically and statistically significant genes from the clinical datasets. •We have proposed a novel hybrid evolutionary approach based on BPSO (4-2) and BBHA for gene selection.•We have tested the effectiveness of BPSO (4-2)-BBHA with three classifiers, on two benchmark and three GEO datasets from NCBI.•The developed approach (BPSO (4-2)-BBHA/SPLSDA) compare with several state-of-the-art methods leads to a better performance.•We have found the optimal subset of genes in each dataset and have used FURIA to find the relation between candidate genes.•Applying BBHA as the local optimizer for BPSO (4-2) helps it to avoid being trapped in a local optimum.