DEVELOPMENT OF A MACHINE LEARNING- BASED MODEL FOR A COMPREHENSIVE ASSESSMENT OF POLYPHARMACYINDUCED DRUG-DRUG INTERACTIONS
Alexander Andreevich Karandeev, N.A. Yashin, A.A. Petrova, K.A. Atroshkin, Nikolay L. Shimanovsky · Medicina i vysokie tehnologii · 2024
Modern algorithms based on Big Data analysis open opportunities to identify drug interactions that are difficult to unveil using only traditional methods of mathematical statistics. This study describes a method of detection of undesirable inter- actions using t-SNE algorithm to optimize search results. Based on the data from the opensource DrugCentral database, using statistical cluster analysis, potential risks of drug interactions used in the pharmacotherapy of cardiac and immune-mediated inflammatory diseases were identified. By reducing the size of the data using the t-SNE algorithm, the clustering process was significantly accelerated and the visualization of interactions was improved. Potential clinically significant interactions were predicted, confirmed by further analysis of the spectrum of clinical effects of individual drugs. The results show a significant improvement in identifying negative interactions and prove high accuracy of the method, which makes it possible to recommend this approach for use in clinical practice to improve safety for patients experiencing polypharmacotherapy.