PROJECT SUMMARY Alzheimer’s Disease (AD) affects millions of Americans, yet there are no treatments that meaningfully affect disease progression once symptoms manifest. This has shifted the focus to early detection and intervention, which is thought by many researchers to offer the best chance of slowing or stopping the progression of AD. However, trials aimed at averting the underlying causes of disease have proven difficult because pathological changes in AD happen well in advance of cognitive decline. A widely-available genetic risk prediction mode (GRPM) for determining AD risk early in life, while prevention might still be possible, would allow early treatment intervention, life planning, enrollment in clinical trials, and improved patient stratification for testing treatment effectiveness. However, despite recent advancements, GRPMs for late-onset AD lack sufficient discrimination ability to support such applications, especially in non-European populations. Given the lack of effective treatments once symptoms have manifested and the socioeconomic consequences at stake, there is a serious unmet need for a widely-available GRPM able to accurately assess a patient’s risk in middle age or earlier, before neurodegeneration begins. To address this need, Parabon has developed a GRPM able to accurately predict an individual’s risk of developing AD at various ages. This model will be commercialized as a direct-to-consumer (DTC) genetic test that can be used by individuals to learn about their future risk and by researchers to recruit and stratify subjects for clinical trials. The Phase II model combines machine learning, a polygenic risk score, and deep learning and achieves state-of-the-art prediction accuracy with an AUC of 0.83 in cross-validation and 0.80 in an independent replication set. However, this model, like most genetic risk scores, was built using only subjects of European descent and thus has reduced accuracy in non-Europeans. The first Aim of this Phase IIB project is to enhance our feature selection and predictive modeling pipeline to detect and utilize genetic variants both across and within ancestral groups. We will implement a novel approach to encoding ancestry information and searching for epistatic interactions between genetic variants on different ancestral backgrounds. The second Aim is to determine how best to communicate the results of the risk prediction to a consumer in a manner that is informative and culturally sensitive while also managing emotional impact, followed by a usability evaluation, engagement with the community, prototyping of the service that will form the foundation of the service that will be made available to customers, and a pilot study.