SBIR-STTR Award

Late-onset Alzheimer's Disease (LOAD) affects millions of elderly people in the United States, yet there are no well-established clinical guidelines for assessing a person's relative risk. Accurate assessment of lifetime risk for LOAD would give high-risk individuals the opportunity to undergo regular biomarker screening for signs of disease and to modify environmental risk factors or participate in prospective clinical trials. This Phase I SBIR project aims to develop a risk prediction model for LOAD that meets or exceeds the accuracy standards established in 1998 by the Working Group for Biochemical and Molecular Markers of Alzheimer's Disease. The recent release of whole-genome sequence data from an extensively phenotyped cohort of the Alzheimer's Disease Neuroimaging Initiative creates a unique opportunity to develop the methodology needed to successfully construct such a risk prediction model. The Parabon team will undertake three specific aims in pursuit of the final goal of producing a risk prediction model that can be used in the clinic. First, a novel methodology will be created for analyzing whole-genome sequence data to discover common SNPs, rare variants, and epistatic interactions that significantly associate with LOAD endophenotypes, the specific physiological changes that underlie disease. This will require innovative algorithm and software development, particularly the implementation of multi-objective optimization in our existing evolutionary search algorithm for detecting epistasis. Second, the discovered significant variants will be built into risk prediction models for each endophenotype using state-of-the-art machine learning methods. These models will then be combined into a single predictive model for lifetime risk, which will be validated in an independent cohort from the Alzheimer's Disease Sequencing Project. Finally, to quantify the confidence associated with each prediction made by the model, algorithms and software for calculating confidence intervals will be developed and implemented. Each new prediction will be presented with a measure of confidence to enable clinicians to determine what, if any, intervention is necessary. When these aims have been completed, Parabon will have produced the first clinically relevant genetic risk prediction model for late-onset Alzheimer's Disease, as well as a suite of software that can be used in the development of other diagnostics. In Phase II, we will move beyond the ADNI-supplied endophenotypes, using image processing and deep learning to infer neuroimaging features most relevant to AD diagnoses, as well as work to validate the predictive models in a larger, more diverse cohort across multiple sites.

Public Health Relevance Statement:


Public Health Relevance:
This Phase I SBIR aims to develop a highly accurate predictive model for lifetime risk of late-onset Alzheimer's Disease to enable early identification of high-risk individuals for participation in clinical trials and regular screening for signs of disease. In pursuit of this goal, the Parabon team will develop algorithms and software to build a novel methodology for the analysis of whole-genome sequence data and endophenotype measures from the Alzheimer's Disease Neuroimaging Initiative. This project will produce the first clinically relevant risk prediction model for late-onset Alzheimer's Disease and a suite of software that can be used for the production of diagnostics for other diseases.

NIH Spending Category:
Acquired Cognitive Impairment; Aging; Alzheimer's Disease; Alzheimer's Disease including Alzheimer's Disease Related Dementias (AD/ADRD); Bioengineering; Biotechnology; Brain Disorders; Dementia; Genetics; Human Genome; Neurodegenerative; Neurosciences; Prevention

Project Terms:
Affect; Algorithmic Software; Algorithms; Alzheimer's Disease; Alzheimer's disease risk; American; base; Biochemical Markers; Biological Markers; Candidate Disease Gene; Cause of Death; Clinic; Clinical; Clinical Trials; clinically relevant; cohort; Computer software; Confidence Intervals; cost; Data; Dementia; Development; Diagnosis; Diagnostic; Disease; disease diagnosis; Early identification; Elderly; endophenotype; Environmental Risk Factor; Etiology; Genetic; Genetic Epistasis; Genetic Risk; Genetic screening method; genome analysis; genome sequencing; genome-wide; genome-wide analysis; Goals; Guidelines; Heritability; Heterogeneity; high risk; image processing; imaging biomarker; improved; Incidence; Individual; innovation; Intervention; Late Onset Alzheimer Disease; Learning; Life; Lifetime Risk; Machine Learning; Measures; meetings; Methodology; Methods; Modeling; molecular marker; neuroimaging; novel; Patients; Persons; Phase; Phenotype; Physiological; predictive modeling; Presenile Alzheimer Dementia; Production; prospective; public health relevance; rare variant; Relative Risks; Reporting; Risk; screening; Sensitivity and Specificity; Single Nucleotide Polymorphism; Site; Small Business Innovation Research Grant; software development; Testing; United States; Validation; Variant; Work; working group

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 test for determining AD risk early in life, while prevention might still be possible, would allow early treatment intervention, enrollment in clinical trials, and improved patient stratification for testing treatment effectiveness. However, despite recent advancements, genetic risk prediction models (GRPMs) for late-onset AD (LOAD) lack sufficient discrimination ability to support such applications. 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 teamed with AD researchers from Washington University and Emory to develop a GRPM able to accurately predict an individual's risk of developing LOAD at various ages. Phase I demonstrated that this GRPM, which exploits diagnostic heterogeneity, non-additive (epistatic) interactions among variants, and machine learning, significantly outperforms traditional risk factors. In Phase II, the GRPM will be further optimized, validated, and commercialized as a direct-to-consumer (DTC) genetic health risk assessment test. In Aim 1, thousands of new and existing case and control subjects with genotypes and detailed phenotypes will be added. In Aim 2, novel approaches to feature selection for epistatic interactions will be implemented to increase the generalizability of selected features. In Aim 3, the selected genomic features will be used to predict imaging and biomarker endophenotypes of LOAD. In Aim 4, the out- of-sample endophenotype predictions will be combined into a final predictive model for AD diagnosis that can be applied to new subjects at any age, which will be validated in an independent replication set. Finally, in Aim 5, data and results will be prepared for scientific publication and submission to the FDA for marketing authorization as a DTC genetic health risk assessment system.

Thesaurus Terms:
Address; Affect; Age; Algorithms; Alzheimer's Disease; Alzheimer's Disease Model; Alzheimer's Disease Risk; American; Amyloid Beta-Protein; Area; Authorization Documentation; Benchmarking; Biobank; Biological; Burden Of Illness; Case Control; Clinical Diagnosis; Clinical Trials; Cohort; Complex; Data; Data Mining; Data Set; Databases; Diagnosis; Diagnostic; Discrimination; Disease; Disease Diagnosis; Disease Heterogeneity; Disease Progression; Disorder Risk; Early Diagnosis; Early Identification; Early Intervention; Early Treatment; Effective Therapy; Endophenotype; Enrollment; Entorhinal Cortex; Framingham Heart Study; Future; Genetic; Genetic Diseases; Genetic Risk; Genetic Screening Method; Genome Analysis; Genome Wide Association Study; Genomic Data; Genomics; Genotype; Goals; Health; Heritability; Heterogeneity; Hippocampus (Brain); Image; Imaging Biomarker; Impaired Cognition; Improved; Improved Outcome; Individual; Intervention; Journals; Knowledge; Late Onset Alzheimer Disease; Life; Life Style; Machine Learning; Manuscripts; Marketing; Methods; Middle Age; Mining; Modeling; Nerve Degeneration; Novel; Novel Strategies; Output; Pathogenesis; Pathologic; Pathway Interactions; Patient Risk; Patient Stratification; Patients; Performance; Persons; Pharmacologic Substance; Phase; Phenotype; Predictive Modeling; Predictive Test; Prevention; Probability; Procedures; Publications; Recovery; Research; Research Personnel; Risk; Risk Assessment; Risk Factors; Sample Size; Sampling; Sensitivity And Specificity; Sex; Small Business Technology Transfer Research; Socioeconomics; Symptoms; System; Tau Proteins; Testing; Training; Treatment Effect; Treatment Effectiveness; Treatment Outcome; United States National Institutes Of Health; Universities; Validation; Variant; Washington; Whole Genome;