Integrated Machine Learning and Pharmacoinformatic Modeling Protocol for the Prediction of Angiotensin II Inhibition in Pulmonary Arterial Hypertension

Abstract

Pulmonary Arterial Hypertension (PAH) is a chronic disease with poor diagnosis and limited therapeutic options. This necessitates the understanding of molecular level information of the therapeutic targets and interventions for the treatment of PAH. Recently, an altered renin angiotensin-aldosterone system (RAAS) has been advocated as a causative factor in PAH pathogenesis. Angiotensin II (Ang II), a key effector peptide of RAAS, initiates inflammatory processes and generates cellular reactive oxygen species mediated pathways resulting in vasoconstriction, proliferation, and inflammation, all of which contribute to PAH development. Angiotensin Converting Enzyme 2 (ACE2), a newly discovered homologue of ACE opposes the detrimental effects of Ang II on the cardio-pulmonary system. However, PAH accounts for an increased Ang II level while down regulating ACE2 and imbalances the overall RAAS mechanism that may stimulate the pathophysiology of PAH. Furthermore, no chemical or therapeutic data of ACE2 activators have been identified. Therefore, the present study aims to explore the binding potential of ACE inhibitors to activate ACE2 for the effective PAH therapy. For this purpose, molecular and predictive modeling approaches with the integration of machine learning (ML) techniques were used to identify the binding mechanistic of ACEIs and to investigate the important pharmacophoric features and selected ML attributes that can be able to differentiate between actives and in-actives ACE2 modulators. Docking analysis was performed to probe the best binding conformation of ACEIs with ACE2. In order to build a 3D predictive model, GRID independent descriptors (GRIND) were used with various sets of conformational inputs. Afterwards, ML techniques were used to develop predictive 2D models for the prediction of ACEI/ACE2 inducers. Docking results validate that ACEIs have the potential to bind and activate ACE2. Furthermore, GRIND model based on 3D docked conformations showed four key features that contribute towards the activity of ligands. Various ML models provided the best accuracy and elucidated the important 2D features. Collectively, these results assist in predicting the behavior of ACEIs along with the important pharmacophoric and 2D features to differentiate between active and in-actives and to predict the new potential modulators of ACE2.