A multiobjective based automatic framework for classifying cancer-microRNA biomarkers

• Two-stage Multiobjective approach for automatic classification
• First stage detects classifier type, parameter and feature-combination
• Second stage combines outputs of first-stage
• Results shown for miRNA and mRNA data sets

Short endogenous RNA aka miRNAs play significant roles in biological processes like RNA silencing and regulation of gene expressions. Several studies have revealed that there might be possible links between oncogenesis and some miRNA expression profiles since profiles of some specific miRNAs are expressed differently in case of normal and tumor tissues. In this paper, a technique based on multiobjective optimization for automatic selection of classifier, its parameters and feature combinations is used for classifying the miRNAs. The proposed approach is divided into two stages. In the first stage, a multiobjective framework in combination with four different classifiers namely Random Tree (RT), Random Forest (RF), Sequential Minimal Optimization (SMO), and Logistic Regression is used. The multiobjective based framework is capable of automatically determining the appropriate classifier, its different parameter combinations and feature combinations for any classification problem. The proposed approach is very generic and can be solved using any multiobjective evolutionary approach. But in the current study the search capability of popular NSGA-II is used. A new encoding strategy is proposed in the current paper to represent all the relevant information (classifier type, parameter combination, feature combination) in the form of a chromosome. Several different mutation operators are developed to accelerate the search process. In the second stage, the best solutions obtained from the first stage are combined using two different approaches: frequency-based approach and a simple ensemble-based approach. The performance of the proposed method has been demonstrated on several real life miRNA and mRNA datasets. Biological relevance of the obtained biomarkers has been reported. Results are compared with several state-of-the-art approaches.