Raheleh Amirkhah - Department of Cell and Molecular Biology, School of Biology, College of Science, University of Tehran, Tehran, Iran
Ali Farazmand - Department of Cell and Molecular Biology, School of Biology, College of Science, University of Tehran, Tehran, Iran
Shailendra K.Gupta - Department of Systems Biology and Bioinformatics, Institute of Computer Science, University of Rostock, Rostock, Germany
Olaf Wolkenhauer - Department of Systems Biology and Bioinformatics, Institute of Computer Science, University of Rostock, Rostock, Germany

The expression of microRNAs (miRNAs) has been changed in most cancers, includingcolorectal cancer (CRC), therefore finding miRNA functions are important for cancermanagement. Emerging evidences have shown that miRNAs have tissue specific functions.Several computational approaches have been developed to predict miRNA targets; however,all of these methods assume a general pattern underlying these interactions and therefore asignificant number of false predictions have also been presented. Current research wasaimed to unravel the most specific features to identify miRNA target interactions in CRC.We developed a novel approach to predict CRC specific miRNA-target interactions using aNaïve Bayes classifier. We trained the algorithm with data from validated miRNA targetinteractions in CRC and other cancer entities. Furthermore, the correlation-based featureselection (CFS) was used to select a set of features that identify CRC-specific miRNAtarget interactions out of 70 features extracted through literature review. Then, thecontribution of each type of feature, i.e. position-based, sequence and structural features,among the selected features was analysed to the performance of the model indiscriminating between functional and non-functional miRNA–mRNA interactions in CRC.The performance of the classifier was evaluated based on the 10-fold cross validation usingreceiver operating characteristic (ROC) curves. The results showed that, though thestructural features ensure a high sensitivity of the model, the sequence features contributetowards the high specificity of the classifier. Additionally, the performance of our modelshowed a significant improvement in comparison to other widely used algorithms based onmachine learning.

MicroRNA, Colorectal Cancer, Target Prediction, Feature, Naïve Bayes, Machine Learning