Computational analysis for drug response prediction and drug repositioning

Abstract

Bio big data analysis plays an important role in studies on drug development by mitigating the constraints posed by their costly and time-consuming nature. Nowadays, the increasing amount of data and the advances in computational analysis methods have facilitated the development of drug research. However, it remains challenging to prepare well-curated datasets with sufficient volume and quality, as well as finding suitable methods to analyze these big data. In the first part of this study, a machine learning model was developed that predicts the pharmacological effects of administering multiple drugs, namely polypharmacy. To address the limitations of existing drug-drug interaction (DDI) prediction models that do not reflect the actual medical state of patients, a gold standard dataset that encompasses patient information, prescription history, and laboratory test results from a publicly accessible clinical database, MIMIC-IV, was first generated. Relevant datasets were systematically preprocessed, and subjected to a suite of machine learning classifiers to predict adverse drug reaction (ADR) signals. In the second part of this study, autoimmune disease drugs, which appeared to be clinically effective for the treatment of COVID-19, were systematically studied for streamlined drug repositioning. Transcriptome data were obtained from patients with COVID-19 and an autoimmune disease, systemic lupus erythematosus (SLE), and were analyzed to identify genes that could be commonly regulated in both diseases. The approaches and insights reaped from this study are anticipated to lay the ground for further drug-related studies.