Relevant and Non-Redundant Feature Selection for Cancer Classification and Subtype Detection

Abstract

Biologists seek to identify a small number of significant features that are important, non-redundant, and relevant from diverse omics data. For example, statistical methods such as LIMMA and DEseq distinguish differentially expressed genes between a case and control group from the transcript profile. Researchers also apply various column subset selection algorithms on genomics datasets for a similar purpose. Unfortunately, genes selected by such statistical or machine learning methods are often highly co-regulated, making their performance inconsistent. Here, we introduce a novel feature selection algorithm that selects highly disease-related and non-redundant features from a diverse set of omics datasets. We successfully applied this algorithm to three different biological problems: (a) disease-to-normal sample classification; (b) multiclass classification of different disease samples; and (c) disease subtypes detection. Considering the classification of ROC-AUC, false-positive, and false-negative rates, our algorithm outperformed other gene selection and differential expression (DE) methods for all six types of cancer datasets from TCGA considered here for binary and multiclass classification problems. Moreover, genes picked by our algorithm improved the disease subtyping accuracy for four different cancer types over state-of-the-art methods. Hence, we posit that our proposed feature reduction method can support the community to solve various problems, including the selection of disease-specific biomarkers, precision medicine design, and disease sub-type detection.

Molecular Cluster Mining of Adrenocortical Carcinoma via Multi-Omics Data Analysis Aids Precise Clinical Therapy

“…The expression of hundreds of transcripts is often cataloged by RNA-seq measurements, but most are redundant (i.e., strongly correlated) or noisy. In addition, the number of samples available is less than the number of features owing to the expenses associated with conducting experiments, which makes it simple for conventional machine learning and statistical algorithms to overfit the biological data [ 71 ].…”

The Identification of Beef and Pork Using Neural Network Based on Texture Features

“…Therefore, a technique is needed to obtain the least overlapping features. The best features are ideally free from redundancy with each other (Rana et al, 2021).…”

“…In contrast with the previous two researchers in this paper, the author conducted multi features selection for multi objects applied to the digital image identification of beef and pork. The key feature is considered relevant when having minimum overlap (Rana et al, 2021). The fundamental problem for overlap feature is that this set of features do not comprehensively represent the value of a feature as a target as it still contains the values of other features.…”

“…The fundamental problem for overlap feature is that this set of features do not comprehensively represent the value of a feature as a target as it still contains the values of other features. The challenge encountered is how to choose one or several of those overlap features as candidates for the best features (Rana et al, 2021). The basic assumption is the smaller range of overlap value of a feature, the better the feature is chosen as the winner of the selection (key feature).…”

ExhauFS: exhaustive search-based feature selection for classification and survival regression

“…, the most differentially expressed genes between two classes of interest) and the “native” selection procedures of the most important features from some models, such as random forests, regression techniques with L 1 -regularization and many others ( Saeys, Inza & Larranaga, 2007 ; Chandrashekar & Sahin, 2014 ; Wang, Wang & Chang, 2016 ). Besides that, several approaches were designed specifically for classification problems involving cancer transcriptomics data, including gene ranking, filtration and combining the most relevant genes in a single model ( Arakelyan, Aslanyan & Boyajyan, 2013 ; Zhang et al, 2021 ; Rana et al, 2021 ).…”