A Machine Learning Algorithm Suggests Repurposing Opportunities for Targeting Selected GPCRs

Repurposing utilizes existing drugs with known safety profiles and discovers new uses by combining experimental and computational approaches. The integration of computational methods has greatly advanced drug repurposing, offering a rational approach and reducing the risk of failure in these efforts. Recognizing the potential for drug repurposing, we employed our Iterative Stochastic Elimination (ISE) algorithm to screen known drugs from the DrugBank database. Repurposing in our hands is based on computer models of the actions of ligands: the ISE algorithm is a machine learning tool that creates ligand-based models by distinguishing between the physicochemical properties of known drugs and those of decoys. The models are large sets of "filters" made out, each, of molecular properties. We screen and score external sets of molecules (in our case- the DrugBank molecules) by our agonism and antagonism models based on published data (i.e., IC₅₀, K_i, or EC₅₀) and pick the top-scoring molecules as candidates for experiments. Such agonist and antagonist models for six G-protein coupled receptors (GPCRs) families facilitated the identification of repurposing opportunities. Our screening revealed 5982 new potential molecular actions (agonists, antagonists), which suggest repurposing candidates for the cannabinoid 2 (CB2), histamine (H1, H3, and H4), and dopamine 3 (D3) receptors, which may be useful to treat conditions such as neuroinflammation, obesity, allergic dermatitis, and drug abuse. These sets of best candidates should now be examined by experimentalists: based on previous such experiments, there is a very high chance of discovering novel highly bioactive molecules.