Using ML to repurpose FDA drugs for the treatment of Diabetic Cardiomyopathy

PARP-1 (Poly ADP Ribose polymerase) functions to repair damage to DNA and is implicated in a variety of diseases including Diabetic Cardiomyopathy (DCM). Unfortunately, there are few treatments for this disease, and the expenses associated with these drugs present barriers to many. With this project, we developed a neural network that was able to distinguish between inhibitors and non-inhibitors of PARP-1 in order to uncover more accessible treatments of DCM. We collected confirmed inhibitors of PARP-1 from PubChem, clustered these compounds, and performed attribute selection. This data was used to develop the neural network which was able to predict inhibitors of PARP-1 with an accuracy of 97% and an AUROC of 0.98. The model was then run on all FDA drugs, and the top 37 predictions were taken. In protein ligand docking simulations, the predicted inhibitors had a significantly better binding affinity for PARP-1 than the control group.