Machine Learning Based Methods and Best Practices of microRNA-Target Prediction and Validation

Abstract

Within the last years, more and more noncoding RNAs (ncRNAs) became the focal point to understand cell regulatory mechanisms because one class of ncRNAs, microRNAs (miRNAs), plays an essential role in translation repression or degradation of specific mRNAs and is implicated in disease etiology. miRNAs can serve as oncomiRs (oncogenic miRNAs) and tumor suppressor miRNAs, thus, miRNA therapeutics in clinical trials have become a vital component with respect to cancer treatment. To circumvent side-effects and allow an accurate effect it is crucial to accurately predict miRNAs and their mRNA targets. Over the last two decades, different approaches for miRNA prediction as well as miRNA target prediction have been developed and improved based on sequence and structure features. Nowadays, the abundance of high-throughput sequencing data and databases of miRNAs and miRNA targets from different species allow the training, testing, and validation of predicted miRNAs and miRNA targets with machine learning methods. This book chapter focuses on the important requirements for miRNA target prediction tools using ML like common features used for miRNA-binding site prediction. Furthermore, best practices for the prediction and validation of miRNA-mRNA targets are presented and set in the context of possible applications for cancer diagnosis and therapeutics.

Keywords: Best practices; Cancer; Machine learning; miRNA validation; miRNA–mRNA target prediction.