Jeffrey J. Saucerman, PhD, FAIMBE, FAHA: No financial relationships to disclose
Description of session (include background & scientific importance): AI and machine learning excel at pattern recognition but rarely provide mechanistic understanding. Here we introduce LogiRx, a mechanistic machine learning method that predicts how drugs modulate cell phenotypes. We applied LogiRx to identify pathways that cause drugs to attenuate cardiomyocyte hypertrophy, a clinical predictor of heart failure. As predicted, escitalopram (Lexapro) and mifepristone both reduced hypertrophy of cultured cardiomyocytes via off-target pathways in subsequent perturbation experiments. Further, escitalopram attenuated cardiomyocyte hypertrophy in mice and patients with depression had a lower incidence of cardiac hypertrophy. These findings validate LogiRx for discovery of new drug pathways, which may enable repurposing of existing drugs and illustrate the value of hybrid mechanistic-machine learning approaches.
Learning Objectives:
Upon completion, participant will know advantages and disadvantages of mechanistic modeling and machine learning.
Upon completion, participant will be able to explain the strengths of combining mechanistic and machine learning approaches.
Upon completion, participant will be able to describe the insights obtained into drug repurposing for heart disease.
Upon completion, the participant will describe other drugs and diseases that may benefit from this approach.
