To stem the rising incidence of Alzheimer's disease (AD) in our aging population, new methods to repurpose and combine drugs against Alzheimer's disease (AD) are acutely required. This is a challenge, however, because the complex polygenic basis of AD remains opaque, and rational methods to repurpose drugs are in early years, even for well-defined gene targets. To address these problems, we propose new algorithms to integrate data on a very large scale so as to combine evolutionary information and high-throughput experimental observations with the knowledge conveyed by text in the literature. First, to detect disease-relevant genome variations in AD patients, Aim 1 will combine a novel mathematical calculus of mutational landscapes with machine learning, in so doing suggesting primary candidate genes for drug targeting based on signs of mutational selection in cases or controls. Next, to repurpose and combine drugs targeting these genes, Aim 2 will map a large fraction of all that is known about genes, phenotypes, and drugs into a single high-dimensional network that represents their interactions as described in various databases (structured data) and in the literature (unstructured data). The topology of this network will determine the optimal choice of single drug or combination therapy in an approach that can be personalized. Finally, to validate efficacy experimentally, Aim 3 will test both our candidate genes and drugs with state-of-the-art in vitro and in vivo screens. Feasibility rests with prior studies on evolution, networks, systems, and text-mining that demonstrate accurate predictions of deleterious mutations and their clinical sequelae and the discovery of drivers of diseases. Broadly, this work will yield proof of principle for a novel quantitative model that integrates fundamental concepts from mathematics and molecular evolution, and for a low resolution but large-scale map of biomedical knowledge in which network notions of distance computed by machine learning identify relevant functional hypothesis that would otherwise be easily overlooked. The result will be a new experimental ability to unravel the genotype-phenotype relationship in Alzheimer's Disease so as to guide drug therapy.
Alzheimer's is a devastating disease projected to rise dramatically in our aging population. To develop drugs against it, we propose to train machine learning on the DNA of patients with Alzheimer's to detect genes with abnormal mathematical features in their mutations. Then, we will place these genes in a ?map?, which represents a large fraction of biomedical knowledge in a high-dimensional network. The drugs that fall near our genes of interest in this map are our candidates, and experiments will test in cell and mouse models whether they affect the hallmark signs of Alzheimer's.
