MATHEMATICAL MODELLING OF MIRNAS INVOLVED IN CANCER

Abstract

miRNAs are small ~21 nucleotide, non-coding, endogenous RNA molecule that has regulatory role in the gene expression of plants and animals .miRNAs regulates gene expression by translational repression, mRNA cleavage, and mRNA decay at post transcriptional level. Recent studies have identified the role of miRNAs in different cancer formations. In addition, miRNAs have been found to function as cancer suppressors. Difference in the expression level of certain miRNAs have been related to the promotion of cancer by negative regulation of tumor suppressor genes. This suggests that miRNAs has potential role in cancer therapy as well as in diagnosis. Therapeutics evaluation of these miRNA requires in vitro and in vivo RNAi based studies. In the current study we have suggested a computational approach to identify these cancer related miRNAs in humans. This approach can be used to skew the dataset of available miRNA for laboratory validation. Further it can be used for prediction of new miRNAs and their role in different cancers. In this approach we have selected all the miRNAs that are known to be involved in cancer from miR2disease database. Then, different features of the miRNA are identified at different developmental stages (pri-miRNA, pre-miRNA and miRNA) and these features (descriptors) were utilized for the specific classification of cancer and/or non-cancer miRNA. Mathematical modelling of miRNAs involved in cancer includes the finding of cancer and non-cancer miRNAs dataset, for positive and negative control. Then, we identified the sequence, structure and the energy based descriptors. Extraction of values of different descriptors on the basic of structural and sequential features. Computer program was developed to extract these values. Mfold were used to predict secondary structure and libraries of RNAfold were implemented in program to calculate the minimum folding energy. Classification algorithms of WEKA software were used to classify the two datasets. Random Forest, MultilayerPerceptron and J48 classification algorithms has shown higher accuracy.