Predicting the potency of anti-Alzheimer drug combinations using machine learning

Abstract

BACKGROUNDClinical trials of single drugs for the treatment of Alzheimer Disease (AD) have been notoriously unsuccessful. Combinations of repurposed drugs could provide effective treatments for AD. The challenge is to identify potentially potent combinations. OBJECTIVETo use machine learning (ML) to extract the knowledge from two leading AD databases, and then use the machine to predict which combinations of the drugs in common between the two databases would be the most effective as treatments for AD. METHODSThree-layered neural networks (NNs) having compound, gated units in their internal layer were trained using ML to predict the cognitive scores of participants in either database, given the other data fields including age, demographic variables, comorbidities, and drugs taken. RESULTSThe predictions from the separately trained NNs were strongly correlated. The best drug combinations, jointed determined from both sets of predictions, were high in NSAID, anticoagulant, lipid-lowering, and antihypertensive drugs, and female hormones. CONCLUSIONThe results suggest that AD, as a multifactorial disorder, could be effectively treated using a combination of repurposed drugs.