Logic-based machine learning predicts how escitalopram attenuates cardiomyocyte hypertrophy

Abstract

Cardiomyocyte hypertrophy is a key clinical predictor of heart failure. High-throughput and AI-driven screens have potential to identify drugs and downstream pathways that modulate cardiomyocyte hypertrophy. Here we developed LogiRx, a logic-based mechanistic machine learning method that predicts drug-induced pathways. We applied LogiRx to discover how drugs discovered in a previous compound screen attenuate cardiomyocyte hypertrophy. We experimentally validated LogiRx predictions in neonatal cardiomyocytes, adult mice, and two patient databases. Using LogiRx, we predicted anti-hypertrophic pathways for 7 drugs currently used to treat non-cardiac disease. We experimentally validated that escitalopram (Lexapro) and mifepristone inhibit hypertrophy of cultured cardiomyocytes in two contexts. The LogiRx model predicted that escitalopram prevents hypertrophy through an off-target serotonin receptor/PI3Kgamma pathway, mechanistically validated using additional investigational drugs. Further, escitalopram reduced cardiomyocyte hypertrophy in a mouse model of hypertrophy and fibrosis. Finally, mining of both FDA and University of Virginia databases showed that patients with depression on escitalopram have a lower incidence of cardiac hypertrophy than those prescribed other serotonin reuptake inhibitors that do not target the serotonin receptor. Mechanistic machine learning by LogiRx discovers drug pathways that perturb cell states, which may enable repurposing of escitalopram and other drugs to limit cardiac remodeling through off-target pathways.