Combined Machine Learning and Pharmacoinformatic- guided Protocol for the Modulation of GPCRs in Migraine

Abstract

G-protein-coupled receptors (GPCRs) are the largest family of membrane proteins that play a fundamental role in cellular signaling. These receptors act as transducers that convey external stimuli to intracellular signaling cascades, thereby regulating a wide array of physiological and pathological processes. In migraine, binding of certain signaling molecules including CGRP, PACAP, adenosine, PGE2 to their specific GPCRs activates the cAMP dependent pathway and activates the pKA. This in turn causes the opening of ATP-sensitive potassium (kATP) channels and large conductance calcium-activated potassium (BkCa) channels. These events result in potassium efflux causing hyperpolarization of vascular smooth muscles, vascular leakage and neurogenic inflammation. These all events result in the sensation of pain associated with migraines. GPCRs impact the cognitive impairment associated with neurodegenerative disorders via intricate signaling pathways including PI3K/Akt, GSK-3β, ERK ½, and CREB which cause neurofibrillary tangles and Aβ plaques formation, synaptic dysfunction and neuronal damage leading to various neurological conditions including Migraine, Alzheimer’s disease, Parkinson’s disease, Huntington’s disease, dementia etc. The prevalence of GPCRs as targets for approximately one–third of the FDA approved drugs underscores their potential role in drug discovery and emphasizes the importance of continued research to develop novel therapies for diseases. Several machine learning models have been developed for classification of GPCR drugs as agonists or antagonists. However, there is no ML model specifically tailored for migraine vs other CNS drugs classification especially for multi-target drug design. For this purpose, integrated molecular modelling and machine learning approach was applied to develop a classification model that differentiates between antagonists that bind to receptors involved in migraine and those involved in other central nervous system diseases. The methodology encompassed the identification of 10 GPCRs associated with migraine and other CNS diseases, followed by their data collection from CHEMBL and preprocessing to compile a dataset with of 1407 antagonists. The docking process employed a “one-to-all” approach, assessing each of the 10 GPCRs against all ligands, resulting in 60,232 unique GPCR-antagonist conformations. Furthermore, proteinligand fingerprints were generated (via a python library ProLIF) for all the docked complexes and a unified binary dataset was generated after undergoing the preprocessing steps in which “1” and “0” represented the presence and absence of interactions. Afterwards, ligands were labelled based xv on their effective binding with GPCRs (Target Label) and disease association (Disease Label). As a ligand can bind to more than one GPCRs effectively, resulting in multiple target and disease labels, multi-label binarization was applied. Finally, multiple ML models were applied in multioutput and multi-step format in order to address the complex task of multi-label classification and prediction of GPCR Target labels and Disease labels (Migraine, Other CNS, Both) within a single model (Drug-Target-Disease relation). Out of these models, XGBoost classifier performed best with an accuracy of 100% on the training set and 98.2% on the testing set. Further, a web application “NeuroLinkMigraine” was developed by the deployment of our best classifier i.e. XGBoost. NeuroLinkMigraine offers a groundbreaking capability to predict the GPCR Targets and Diseases associated with a ligand simultaneously. In conclusion, our classifier and web application study can help in the simultaneous screening of antagonists against multiple GPCRs involved in migraine and other CNS diseases, providing insights into polypharmacology, combination therapies and drug repurposing.