Integrating tertiary structural representations for miRNA - drug interaction prediction

Abstract

miRNAs are small non-coding RNAs that regulate gene expression and are dysregulated in many diseases. Its aberrant expression led to an interest in therapeutic intervention, where it can target drugs that are undruggable by proteins by acting as a gene regulator. Thus, exploring potential drugs from their aberrant miRNA information may facilitate novel treatments to tackle the current disease complexity. With the long process of wet-lab screening, it is imperative to provide a rapid miRNA and drug interaction screening to accelerate the drug discovery processes. This study proposes MDITransNet, a miRNA-drug interaction (MDI) neural network model that incorporates structural information of miRNA and molecular graphs of drugs that is trained under an adapted transformer block module of structure prediction. MDITransNet was able to retain a notable performance than conventional machine learning models with an area under the receiving operating curve (AUC) of 92% and 87% on the non-redundant test set, hairpin, and mature, respectively.