ARTIFICIAL INTELLIGENCE (AI) APPROACHES IN DRUG DISCOVERY: DEVELOPMENT AND VALIDATION OF NEW STRATEGIES IN VIRTUAL SCREENING CAMPAIGNS

The field of drug discovery is increasingly benefitting from the advent of artificial intelligence (AI), which has proven to enhance both the efficiency and accuracy of Computer-Aided Drug Design (CADD), while significantly reducing the associated costs and timelines. This Work examines various case studies in which AI has been applied, with the general purpose of improving the accuracy and reliability of predictive studies. The first analysed area is the xenobiotics metabolism, assessed with the support of highly accurate input data, collected from the internally curated MetaQSAR database. The glutathione conjugation was chosen to investigate the enhancing role of structure-based studies in predicting the occurrence of a single metabolic reaction with Random Forest classification models. Later, a new tool (MetaSpot) able to predict the site of metabolism for most metabolic reactions was presented, as the natural extension of the previously published MetaClass, since they demonstrated synergistic and complementary roles. As a second application, XGBoost regression models were employed to investigate the polyspecificity of ABCB transporters, aiming to overcome common drawbacks in QSAR studies. Finally, given the proven relevance of descriptors in supervised predictive modelling, novel kinds of structure-based descriptors were developed and validated. FingerInt and HyperVolume are here proposed for enriching the pool of docking-derived descriptors, avoiding the multitude of ligand chemical descriptors, far away from the purpose of Explainable Artificial Intelligence (XAI).