A knowledge map to find Alzheimer's disease drugs

ABSTRACT. This Supplement extends Aims 1 and 2 of the parent grant on Alzheimer's Disease (AD) bydeveloping: prospective benchmarks for algorithms that predict biomarkers of disease risk (Aim 1) and newalgorithms to support drug repositioning (Aim 2). Both extensions strengthen Aims 1 and 2 for AD but also haveimmediate applications for research on COVID-19 disease in keeping with NOT-AG-20-022.AIM 1 of the parent grant develops EA-ML, a Machine Learning (ML) pipeline to compare coding mutations inindividuals with and without AD. The output is a list of genes with which to predict AD risk from their mutations.While the parent grant has multiple criteria for success, none are prospective given the vast lead-time betweenAD onset and symptoms. Supplemental Aim 1 adds prospective testing, using COVID-19. This is possiblebecause the UK Biobank has begun to annotate its 50,000 public exomes with the COVID-19 status ofindividuals, including who had severe morbidity or mild symptoms at worst. The biobank will also add 150,000more exomes by end 2020. Accordingly, we will apply EA-ML to the current UK biobank data to identify humangenetic biomarkers that distinguish severe from mild cases and then test EA-ML predictions of COVID-19virulence prospectively, on the exomes that are newly added to the biobank. As a further new benchmark, wewill also compare EA-ML to a novel "EA-Wavelet" algorithm, also tested prospectively on COVID-19. EA-Wavelet sorts cases from controls by factoring EA over the entire network of human protein-protein interactions.The results will tell us which of EA-ML, EA-Wavelet, or combination thereof is the best at identifyingcritical biomarkers and clinical risk of AD, while also doing the same for COVID-19.Aim 2 of the parent grant develops drug repositioning for AD by linking target genes and drugs via knowledgemaps of functional interactions. Here, we propose a complementary approach that connect genes to drugsvia structural maps of binding epitopes. For this we will comprehensively map evolutionarily important sitesin the structural proteome of genes that are associated with AD. The approach exploits EA theory to measurepast and present evolutionary forces in fitness landscapes, and it takes into account current sequence variationsto guard against any possible mutational escape from drugs that target these epitopes. The output will be surfaceaccessible regions of proteins that can then be used for (i) computational docking of small molecules towardsdrug repurposing, combination therapy, and lead discovery for drug design3-5; (ii) engineering mimetic peptidesor other molecules that can inhibit normal interactions6; and (iii) CRISPR engineering or peptide synthesis thatcreate antigens for more effective vaccines7, 8. These automated mapping tools are general, and besides inSARS-CoV-2, will identify an entire new structural library of functional sites to target for AD therapy withrepurposed drugs.