A stacked autoencoder based gene selection and cancer classification framework. (September 2022)

• Record Type: Journal Article
• Authors: Gokhale, Madhuri; Mohanty, Sraban Kumar; Ojha, Aparajita
• Is Part Of: Biomedical signal processing and control. Volume 78(2022)
• Publication Date: 2022-09
• Subjects: Gene selection -- Cancer classification -- Stacked autoencoder; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103999
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Cancer is one of the most common causes of death worldwide and is, therefore, a prominent area of biomedical research. Cancer is a genetic disease in which improperly functioning genes tend to change expressions. Thus, gene expression analysis is utilized for early diagnosis of cancer prognosis, and therapy prediction in a clinical environment. Usually, some dominant genes among thousands of them play an important role in the diagnosis of cancer. But designing a suitable framework to find out the key set of genes is a challenging task. Numerous gene selection approaches have been introduced by researchers for cancer classification, using statistical, or traditional feature selection methods. In recent years, deep learning methods have also been applied for gene selection using autoencoder networks. However, improving the accuracy of cancer classification still remains a challenging task. In the present paper, a stacked autoencoder-based framework is proposed for gene selection and cancer classification. Nine different classifiers are employed to evaluate the performance of the gene selection model. Then the best performing combination of gene selection and cancer classification models are chosen to finally select the genes. Random Forest and Support Vector Machine show better performance on ten different benchmark datasets, when the gene selection is done using the stacked autoencoder. The classifier with the highest accuracy is selected to build the cancer classification model. The proposed model outperforms seven existing methods on all the ten datasets. Highlights: A framework for gene selection and cancer classification is proposed that employs an autoencoder and a classifier from a bucket of nine classifiers. Selection of the autoencoder and related hyper parameters are done by an empirical study on ten different gene expression datasets. The number of most prominent genes is decided by iteratively decreasing the number of genes and finding the least possible number for effective cancer classification. The model selected through the proposed framework outperforms seven other existing models on all the ten datasets. Biological significance of the results has also been analyzed.