Journal of Health and Biomedical Informatics

Introduction: Prediction of MicroRNAs (miRNAs) targets has a major importance. Development of calculating methods and also cost-effective and time-saving laboratory researches, has great effect on the production of therapeutic medicines like anti cancerous drugs. Since miRNAs have been recently identified, the researches trend was slow at first but by development of biologic databases and understanding its importance, scientists enhanced the speed of studies and paid more attention to this field. Until now, several computerized methods have been developed for prediction of microRNAs (miRNAs) targets but most of these methods have high false positive and further studies are required to improve these methods. Since, recent studies show that miRNAs have different targets in several tissues, this study aimed to provide a computerized method for prediction of miRNAs target in breast cancer.

Method: In this study, at first, all types of features were extracted, then, dominant features were selected via CFS and Relief methods. Smart models such as nervous network, Support Vector Machine with three different cores, Naïve Bayes algorithm and Random Forest decision tree using ten cross accreditation method were tested and its results compared and analyzed. In order to validate the results, gene expression profiling was used.

Results: Analyzing miRNA and gene expression profiles, the classifier predicted 124 functional interactions involving 21 miRNAs and 38 mRNAs in breast cancer.

Conclusion: In terms of bioinformatics, this approach was validated for breast cancer but for further validation, experimental methods also should be used.