Cognitive Computing of Alzheimer’s Disease Genes and Risk The molecular basis and genetic architecture of dementia remain a puzzle. As no drug yet prevents, delays, or reverses it, aging populations potentially face a tidal threat of incipient and socially disruptive Alzheimer’s Disease (AD) cases. Genome-wide association studies (GWAS) have linked over 100 loci with AD and explain much of population attributable risk, but only a fraction of heritability. This heritability gap means it remains difficult to design and assess which surveillance, screening, preventive, and stratification programs are effective. In turn, this hinders therapeutic trials. The challenge in translating genetic variants into patient classifications is twofold. First, AD is polygenic, so relevant disease driving mutations are spread thin across a multitude of different genes and patients. Second, current interpretations of the deleterious effects of mutations lack accuracy, so the impactful few cannot be distinguished from the benign multitude in any given subject. These problems compound and fog the statistical genetics of AD risk and morbidity with poor signal to noise ratio. The crux of our solution is to add a massive amount of new information, exploit it efficiently through computation, then perform rigorous multi-pronged experimental validation. We start from the hypothesis that AD arises through mutational perturbations that affect functional pathways beyond the built-in evolutionary tolerances. New algorithms compute these excessive mutational forces and place them in integrative machine learning frameworks to sort between AD patients and controls, and which can also reflect functional interactions among proteins or genes. Innovations include a mathematical model of evolution based on calculus; ensemble machine learning over human genome variations; and harmonic analysis of mutational perturbations in functional networks. The outcome will, for the first time, integrate genomic variations relevant to AD in the context of all relevant evolutionary history and all known functional interactions. In practice, this will increase power and resolution, enable gender-specific analysis and AD stratification of men and women, and identify new and experimentally validated AD genes. To carry out this program, AIM 1 will fuse a novel mathematical analysis of evolution with machine learning and network wavelet theory. This will yield complementary integrative approaches to identify genes and mutations that sort AD vs healthy subjects based on the abnormal mutational burden of rare gene variants in sequenced cohorts. AIM 2 will focus similar tools on patients and controls with known paradoxical phenotypes that run counter to their APOEɛ2/4 status. The results will identify modifier genes that drive AD in APOEɛ2 carriers or that protect APOEɛ4 carriers from AD. AIM 3 will provide direct experimental validation, leveraging high-throughput, robot-assisted genetic modifier screening i...