A Drug-Virus Prediction Model Based on an Ensemble Classifier with Feature Optimization: A Case Study with COVID-19

Effective therapeutics are still unavailable for the contagious COVID-19, which originated in Wuhan in December 2019. It is induced by the newly emerged severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). This worldwide outbreak has taken many lives and disordered the world economy. Even though COVID-19 vaccines helped to boost immunity against the virus, an effective medication is still needed to curtail the disease outbreak. Drug repositioning is the process of investigating and finding novel uses for existing drugs in the market. It is an economical and time-saving solution to fight against this rapid and widespread infection. This work, ESRVDA aims to explore and predict Drug-Virus Associations to rank drugs against COVID-19 through an ensemble of Extreme gradient boosting classifier with Synthetic minority oversampling technique (SMOTE) and feature optimization by Recursive feature elimination cross-validation (RFECV). We propose a model to predict drugs for COVID-19 and other viral transmittable diseases through a machine learning-based method. We constructed feature vectors from the chemical structure similarity of drugs and the genomic sequence similarity of viruses. The optimized augmented samples are classified based on the ensemble classifier, and drugs are ranked based on the attained association probabilities. We conducted fivefold cross-validation to assess the performance of the model. The achieved AUC score is 0.9087. Case studies and experimental results and prove the efficacy of the model in the drug-virus association prediction.