Integrating fuzzy rough set-based entropies for identifying drug-resistant miRNAs in cancer

MicroRNAs (miRNAs) are key biomarkers in cancer diagnosis and treatment. Identification of drug-resistant miRNAs may help in effective treatment of cancer. Two new z score based fuzzy rough relevance and redundancy entropies are developed and then a weighted framework is introduced to integrate the entropies for ranking and selecting miRNAs in classifying control and drug resistant patients. Here, two key components of soft computing, fuzzy set and rough set are utilized. The methodology is called a weighted framework for integrating fuzzy rough set-based relevance and redundancy entropies (WFIFRRRE). The z score is used to compute the fuzzy membership of expression values required for both entropies. Fuzziness deals with the overlapping nature of miRNA expression profiles and rough set helps in determining the exact class size. The weights in WFIFRRRE, assigned to relevance and redundancy entropies, are determined in a supervised manner to maximize the $F$ score used for validating the classification performance in discriminating the control and drug-resistant patients. The weights are varied from 0 to 1 in steps of 0.01 which enables an integration between relevance and redundancy entropies. A subset of miRNAs is selected from the ranked list and the performance is evaluated using three benchmark classifiers on eight drug-resistant cancer datasets. Experimental results show that WFIFRRRE provides better prediction accuracy than the popular methods used for comparison. The classification accuracy in terms of $F$ score, achieved by WFIFRRRE, ranges from 0.74 to 1.0, 0.75 to 1.0, and 0.73 to 1.0 using random forest, Naive Bayes, and linear SVM classifiers, respectively. The resultant set of miRNAs obtained using WFIFRRRE is also verified with the help of existing biological studies. The source code of WFIFRRRE is available at https://www.isical.ac.in/ shubhra/WFIFRRRE.html.