Ligand-Based Approach for Multi-Target Drug Discovery: PTML Modeling of Triple-Target Inhibitors

BACKGROUND: Cancers are complex multi-genetic diseases that should be tackled in multi-target drug discovery scenarios. Computational methods are of great importance to accelerate the discovery of multi-target anticancer agents. Here, we employed a ligand-based approach by combining a perturbation-theory machine learning model derived from an ensemble of multilayer perceptron networks (PTML-EL-MLP) with the Fragment-Based Topological Design (FBTD) approach to rationally design and predict triple-target inhibitors against the cancerrelated proteins named Tropomyosin Receptor Kinase A (TRKA), poly[ADP-ribose] polymerase 1 (PARP-1), and Insulin-like Growth Factor 1 Receptor (IGF1R). METHODS: We extracted the chemical and biological data from ChEMBL. We applied the Box- Jenkins approach to generate multi-label topological indices and subsequently created the PTML-EL-MLP model. RESULTS: Our PTML-EL-MLP model exhibited an accuracy of around 80%. The application FBTD permitted the physicochemical and structural interpretation of the PTML-EL-MLP model, thus enabling a) the chemistry-driven analysis of different molecular fragments with a positive influence on the multi-target activity and b) the use of those favorable fragments as building blocks to virtually design four new drug-like molecules. The designed molecules were predicted as triple-target inhibitors against the aforementioned cancer-related proteins. CONCLUSION: Our study envisages the capabilities of combining PTML modeling with FBTD for the generation of new chemical diversity for multi-target drug discovery in oncology research and beyond.