Дослідження методів програмної реалізації оцінки властивостей складних структур

Abstract

This work is dedicated to validating QSAR models, which are pivotal for predicting molecular activity based on structural attributes, with a particular focus on ADMET. It meticulously examines the performance of two prominent methodologies: chemprop, employing graph neural networks for biological activity prediction, and a SVM classification model utilizing molecular fingerprints. Diverse mathematical approaches are rigorously assessed, employing metrics such as balanced accuracy (BA) and area under the ROC curve (AUC). The ultimate objective of this research is to enhance the efficiency of evaluating molecular properties, offering promising implications in the realm of drug discovery.