- Researchers at Massachusetts General Hospital are using machine learning to screen existing drugs—both approved and investigational—for their therapeutic potential in Alzheimer's disease (AD) before they are entered into clinical trials
- In this study, the machine learning framework ranked 80 kinase inhibitors by how well their mechanisms of action were able to predict the correct stage of AD severity from mRNA expression profiles in brain specimens
- JAK, ULK and NEK kinases were among the most consistent targets of top-ranked drugs
- These kinase families are known to target proteins in signaling networks regulating innate immunity, autophagy and microtubule dynamics, which are previously unexplored pathways for AD therapeutics
- Future studies of the drug hits will need to validate the gene targets in cell-based and animal models using CRISPR or another gene-editing technique
The mechanisms underlying Alzheimer's disease (AD) are still poorly understood, and about 200 clinical trials of novel therapeutics have failed. To avoid this substantial loss of time and resources, a better approach might be to repurpose drugs already approved by the FDA.
The traditional approach to drug repurposing is simply to conduct a clinical trial of the existing drug in the new indication, perhaps at a different dose or in different formulation. Researchers at Massachusetts General Hospital suggest a different strategy: use repurposing as a way to test a therapeutic concept. If an approved drug is successful, it could enter a clinical trial directly; otherwise, the proof of concept would accelerate creation of a new molecular entity.
To screen approved and unapproved drugs for their therapeutic potential in AD, Steve Rodriguez, PhD, instructor in Neurology, MassGeneral Institute for Neurodegenerative Disease and Neurologist Mark W. Albers, MD, PhD, of the Department of Neurology at Mass General; Artem Sokolov, PhD, director of Informatics and Modeling at Harvard Medical School; and Clemens Hug, of the Harvard Program in Therapeutic Science; and colleagues have created a novel artificial intelligence framework, which they call Drug Repurposing in Alzheimer's Disease (DRIAD). They describe it and its initial results in Nature Communications.
Background on DRIAD
DRIAD is based on machine learning, a branch of artificial intelligence in which computer algorithms make inferences from large datasets and learn to identify patterns through practice and repetition without being specifically programmed. In this study, DRIAD quantified associations between two sets of inputs:
- mRNA expression profiles from autopsied human brains at various stages of AD progression
- Lists of genes that are differentially expressed when neuronal cells are exposed to a test panel of drugs
For the test panel, the researchers used 80 kinase inhibitors, the largest class of targeted drugs currently available. The panel included three types of drugs: FDA-approved agents, unapproved drugs whose toxicity has been tested in clinical trials and unapproved drugs that so far have been tested only preclinically.
DRIAD ranked compounds by how well their mechanisms of action (as represented by the list of gene names) were able to predict the stage of AD severity. Supplemental material to the article shows how all tested drugs performed.
JAK, ULK and NEK kinases were among the most consistent targets of top performers. These kinase families are known to target proteins in signaling networks regulating innate immunity, autophagy and microtubule dynamics, which are previously unexplored pathways for AD therapeutics. The results also suggested that a successful AD therapeutic might require polypharmacology with respect to these kinases.
To evaluate whether a drug hit from DRIAD affects AD pathophysiology, the next step will be to validate the target in cell-based and animal models using CRISPR or another gene-editing technique. Of course, any drug used in AD will have to cross the blood–brain barrier.
Other research directions could be to run DRIAD in: (a) different subgroups of AD patients, defined by more detailed clinical or pathologic phenotypes, with the aim of developing personalized interventions, and (b) studies that include age-matched healthy individuals, in hopes of learning how to predict AD pathology.
Learn more about research in the Department of Neurology
Learn more about Mass General Neuroscience