An Integrative Machine Learning Framework for Drug Repurposing Using Molecular Docking and Drug Sentiment Analysis in Type 2 Diabetes

Abstract

Type 2 diabetes (T2D) is a major global health challenge that requires continuous investigation of alternative therapeutic strategies. Drug repurposing is a strategic technique that involves identifying new therapeutic applications from existing drugs. This research developed a machine learning framework that combines drug-review sentiment analysis and molecular docking to examine the interaction of selected non diabetic drugs with a type 2 diabetic protein target, and determined their potential to treat T2D under further investigations. Drug review data were obtained from the UCI machine learning repository. The data contains 4,143 records and were divided into training (containing 3,107 records) and testing (containing 1,036 records) sets. Using Term Frequency–Inverse Document Frequency (TF-IDF), textual information from the drug review datasets were converted into numerical features, suitable for model training. Three (3) different supervised learning machine learning algorithms; Logistic Regression, Linear Support Vector Machine, and Naive Bayes classifiers, were evaluated using five-fold cross-validation based on accuracy, precision, recall, F1 score, and macro-F1. Of the three (3) models, Logistic Regression therefore had the best overall performance, achieving a macro recall of 0.567231 and a macro-F1 score of 0.558643, and was subsequently used for the drug-level sentiment analysis. The three non-diabetes drugs, bupropion, esomeprazole, and adapalene, with the highest sentiment scores were selected for molecular docking against dipeptidyl peptidase-4 (DPP-4). Also, Sitagliptin was selected as the control drug and docked against DPP-4. Using CB-Dock2, the predicted docking scores were −6.3 kcal/mol for bupropion, −8.8 kcal/mol for esomeprazole, −10.8 kcal/mol for adapalene, and -9.3 kcal/mol for Sitagliptin. Adapalene demonstrated the best predicted binding score, which suggests that adapalene may have better interaction with DPP-4 than Sitagliprin (control drug). This may require further investigations. The findings demonstrate the potential of integrating patient-generated textual evidence with structure-based molecular analysis to prioritize existing drugs for further investigation in T2D. However, the computational findings do not establish antidiabetic efficacy or clinical suitability and require further validation through additional computational and experimental studies.