Gene Selection for Microarray Cancer Classification based on Manta Rays Foraging Optimization and Support Vector Machines

• Published in: Arabian journal for science and engineering (2011) 2022-02, Vol.47 (2), p.2555-2572
• Main Authors: Houssein, Essam H., Hassan, Hager N., Al-Sayed, Mustafa M., Nabil, Emad
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Cancer; Classification; Colon; Datasets; DNA chips; Engineering; Genes; Heuristic methods; Humanities and Social Sciences; multidisciplinary; Optimization; Optimization techniques; Redundancy; Research Article-Computer Engineering and Computer Science; Science; Support vector machines
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-021-06102-8

In DNA microarray applications, many techniques are proposed for cancer classification in order to detect normal and cancerous humans or classify different types of cancers. Gene selection is usually required as a preliminary step for a cancer classification problem. This step aims to select the most informative genes among a great number of genes, which represent an important issue. Although many studies have been proposed to address this issue, they lack getting the most informative and fewest number of genes with the highest accuracy and little effort from the high dimensionality of microarray datasets. Manta ray foraging optimization(MRFO) algorithm is a new meta-heuristic algorithm that mimics the nature of manta ray fishes in food foraging. MRFO has achieved promising results in other fields, such as solar generating units. Due to the high accuracy results of the support vector machines (SVM), it is the most commonly used classification algorithm in cancer studies, especially with microarray data. For exploiting the pros of both algorithms (i.e., MRFO and SVM), in this paper, a hybrid algorithm is proposed to select the most predictive and informative genes for cancer classification. A binary microarray dataset, which includes colon and leukemia1, and a multi-class microarray dataset that includes SRBCT, lymphoma, and leukemia2, are used to evaluate the accuracy of the proposed technique. Like other optimization techniques, MRFO suffers from some problems related to the high dimensionality and complexity of the microarray data. For solving such problems as well as improving the performance, the minimum redundancy maximum relevance (mRMR) method is used as a preprocessing stage. The proposed technique has been evaluated compared to the most common cancer classification algorithms. The experimental results show that our proposed technique achieves the highest accuracy with the fewest number of informative genes and little effort.