MiRNA-Disease Associations Prediction Based on Improving Feature Vectors Quality Combined with Highly Reliable Negative Samples Selection

Abstract

MicroRNAs (miRNAs) are tiny and non-coding RNA molecules with ~22nt in length. They have important roles in various biological processes, gene expression and influence cellular functions . Determining the miRNAs involved in disease processes is a complex endeavor, hindered by the high expenses and protracted nature of experimental techniques. As a result, reliable computational models are sought as an effective alternative to infer associations between miRNA and disease-related. In this paper, focusing on improving feature vector quality combined with highly reliable negative sample selection, we propose a model to predict miRNA and disease association (MDAs). First, it integrates many miRNAs and diseases similarities into two integrated ones by employing the Fast Kernel Learning (FKL) model, which enhances biological knowledge and minimizes reduce prediction bias. Second, this method uses pair importance obtained by structural Hamiltonian information in the feature vector selection phase to construct quality feature vectors. Third, it solves the unbalanced dataset in machine learning (ML) approaches by using the strategy of getting highly reliable negative samples. Finally, its performance achieves the highest AUCs and AUPRs of 5-fold cross-validation (5-fold-CV) in compared models as well as good validated effectiveness in case studies.