Gene reduction and machine learning algorithms for cancer classification based on microarray gene expression data: A comprehensive review

Highlights

• A comprehensive review on microarray gene expression reduction is proposed.

• A novel taxonomy of up-to-date gene reduction algorithms is presented.

• The strength and limitations of each type of gene reduction algorithms are provided.

• Provide an outline of the accurately used type of machine learning algorithms.

• Analyze the performance of the state-of-the-art algorithms.

Abstract

Disease diagnosis and prediction methods in biotechnology and medicine have significantly advanced over time. Consequently, analyzing raw gene expression is crucial for identifying diseases such as cancer. Interestingly, microarrays are a tool that records gene expression from deoxyribonucleic acid (DNA) or ribonucleic acid. This technique exhibits intriguing characteristics, such as generating high-dimensional data with a small sample size. However, in the case of such dataset, the classification model is prone to overfitting. This limitation can be overcome by reducing the dimensions of the microarray datasets to a reasonable number. Machine learning (ML)-based data reduction has recently achieved considerable attention in genomic research. Therefore, this review examines recent studies that present state-of-the-art data reduction and classification algorithms for microarray gene expression data to diagnose tumors and analyzes their performance. To the best of our knowledge, this is the first review that provides a comprehensive view of data preprocessing, dimensionality reduction, including feature (i.e., gene) selection, feature extraction, and their hybrid, and ML algorithms. The paper is structured as follows. First, this review summarizes several data preprocessing methods applied to gene expression datasets. Then, a detailed review of various ML-based feature selection algorithms, including filter, wrapper, embedded, ensemble, and hybrid algorithms, is discussed. These algorithms are examined under three main classes—supervised, unsupervised, and semisupervised ML. Next, the feature extraction and hybrid of feature extraction and selection algorithms are thoroughly reviewed. Furthermore, a detailed review of broadly applied ML algorithms to simplify tumor and nontumor classification using microarray datasets is presented. Finally, the challenges and open questions related to gene expression datasets for accurate cancer classification and detection are highlighted.

Introduction

In recent years, data analytics have gained considerable attention in the field of biomedical research, with a deep concentration on the most critical clinical conditions, such as cancer (Momeni, Hassanzadeh, Abadeh, & Bellazzi, 2020). For example, DNA microarray techniques have been adopted to represent and describe thousands of gene expressions and generate microarray gene expression data (Sayed, Nassef, Badr, & Farag, 2019). Gene expression profiling expresses the current physiological state of a cell and contains valuable information about gene activity. These data have effectively been used for early diagnosis and prognosis of cancer types, in the past few years (Bolón-Canedo & Alonso-Betanzos, 2019).

A significant defect in such data is that they comprise inordinate gene counts (the curse of dimensionality) against a few observations (the curse of data sparsity). However, high-dimensionality data that include irrelevant, redundant, and noisy genes cause incorrect diagnoses owing to the few cancer-causing genes (Almugren & Alshamlan, 2019c). Such datasets tend to overfitting, producing non-reproducible results and are impacted by high variance (Momeni et al., 2020). For data analysis, coping with a dataset consisting of a limited number of observations and a huge number of genes is a daunting task.

Dimensionality reduction algorithms, including feature extraction and feature selection algorithms, are commonly used to solve this problem. Feature (gene) extraction algorithms are statistical methods that transform an existing input surface into a new one by generating new features based on the extant ones to reduce input features representation (Momeni et al., 2020). Therefore, as they affect the input feature space during the transformation, the final prediction model from this new-feature surface lacks interpretability. Alternatively, in dimensionality reduction, feature selection algorithms are frequently preferred to feature extraction algorithms for microarray gene expression dimensionality reduction (Momeni et al., 2020). These algorithms have been applied to gene expression datasets to select ideal discriminating genes called biomarkers (Zawbaa, Emary, Grosan, & Snasel, 2018). Accordingly, a successful gene selection algorithm should select the fewest genes that achieve the best performance based on specific criteria, such as model accuracy, sensitivity, and sensibility (Chen, Meng, & Su, 2020).

Fig. 1 illustrates the main stages in microarray data analysis. These stages include data exploration, data preprocessing, dimensionality reduction, machine learning (ML) algorithm, and evaluation. As shown in the diagram, data acquisition is the process of obtaining or gathering data to work with. Then, data exploration is used to understand the nature of the acquired data by comprehending its features, format, and quality. Next, data preprocessing is used to convert raw data into a readable and high-quality format. Afterward, dimensionality reduction algorithms can be used to identify the relevant genes. Then, an ML algorithm is used to create a cancer classification model. Finally, the ML model is evaluated.

Surveys that address numerous feature reduction algorithms for gene analysis have been reported in the literature. However, to the best of our knowledge, none provide a comprehensive/systematic and broad view of data preprocessing methods, data reduction algorithms, including feature selection, feature extraction, and their hybrid from the perspectives of an ML and the ML algorithms employed. For example, a previous survey (Kumar, Sooraj, & Ramakrishnan, 2017) focused only on various gene selection algorithms based on supervised learning algorithms. In contrast, another survey (Almugren & Alshamlan, 2019c) concentrated only on hybrid gene selection algorithms. Furthermore, a previous review (Mahendran, Vincent, Srinivasan, & Chang, 2020) focused only on gene selection algorithms from an ML perspective, classifying them into supervised, unsupervised, and semisupervised feature selection algorithms. In contrast, another survey (Hira & Gillies, 2015) presented in 2015 focused only on feature selection and extraction algorithms in DNA microarray datasets.

The present review bridges the gap identified above while examining various recent studies in genomics in detail, investigating the severe difficulty of dealing with microarray datasets —particularly gene expression datasets—and several proposed solutions. A summary of various data preprocessing methods for gene expression datasets is presented. A novel taxonomy of data reduction algorithms is proposed, classifying the reviewed research into three groups: research on feature selection, feature extraction, and their hybrid. Feature selection studies include filter, wrapper, embedded, ensemble, and hybrid algorithms. Therein, each class is further classified as supervised, unsupervised, and semisupervised based on the applied ML algorithm (Fig. 2). Finally, this review examines frequently used ML algorithms and summarizes the deductions and new trends in disease diagnosis and prognosis techniques based on microarray gene expression data.

Key contributions of this review are as follows:

• Examine up-to-date data reduction algorithms for microarray gene expression datasets under three classes: feature selection, feature extraction, and hybrid of feature extraction and selection.

• Examine recent feature selection algorithms for microarray gene expression datasets under five classes: filter, wrapper, embedded, ensemble, and hybrid. They are further classified on the basis of ML algorithms under three classes: supervised, unsupervised, and semisupervised

• Examine ML algorithms used to diagnose cancer based on a DNA microarray.

• Provide a summary of datasets used to develop cancer classification models based on the microarray.

• Analyze the performance of developed algorithms and provide an outline of strengths and limitations of each type of data reduction algorithm.

• Outstanding challenges of coping with microarray datasets and future works are listed and explained briefly, which might help draw a roadmap for researchers on recent trends and potential future research directions.

The rest of the paper is structured as follows. Section 2 presents the methodology and strategies used to obtain study details such as research questions, keywords, and study selection criteria. The data preprocessing overview is thoroughly explained in Section 3. Sections 4, 5, and 6 present the reviews of feature selection algorithms, feature extraction algorithms and the hybrid of feature extraction and selection algorithms, and ML algorithms, respectively. The main observations are presented in Section 7. Finally, Section 8 discusses the challenges and future trends, and Section 9 presents the conclusion of this review.