Center for Alzheimer's and Related Dementias (CARD): Harmonized Data-Derived Resources for the Alzheimer's Disease and Related Dementias Community

National Institute On Aging

Investigators

Linked publications, trials & patents

Abstract

Executive Summary The ADRD Data Harmonization Project represents a transformative initiative within CARD's scope of work (AG000534-03) that has revolutionized scientific investigation through the development of cutting-edge data harmonization tools and methodologies. This comprehensive project has established NIH-leading open science standards for data and code distribution, built automated end-to-end pipelines for multi-omics and genetics analysis, and created innovative AI-powered tools that have significantly accelerated research productivity across the Alzheimer's disease and related dementias research community. The project's success is demonstrated by over 110 publications on PubMed since 2023, the development of multiple publicly available tools including GenoML (accessed by 6.9K+ users), and the establishment of collaborative partnerships with major biobanks and research institutions worldwide. The team's pioneering work in human-in-the-loop AI tooling and federated data harmonization has positioned CARD as a leader in biomedical data science, with significant contributions to NIH-wide AI standardization efforts. Project Summary Narrative This aspect of CARD's scope of work (AG000534-03 Harmonized Data-Derived Resources for the Alzheimer's Disease and Related Dementias Community) has been critical to the Center's success. Every member of the Advanced Analytics team has contributed to these efforts that empower both the team and other teams at CARD. Central to this effort is: (1) establishing NIH-leading open science standards for data and code distribution for transparency, reproducibility, and scalability, (2) building automated end-to-end pipelines for multi-omics and genetics that result in rapid harmonization of data for increased interoperability and more powerful analyses while reducing activation energy, and (3) curating data at both the participant and summary annotation level to maximize the value of these resources, ensuring easily actionable research. These efforts have been crucial to our success, and the over 110 publications on PubMed since 2023 attributable to this scope of work attest to its success. Our team's safe but innovative leveraging of human-in-the-loop AI tooling has allowed us to maximize productivity for all aspects of the core data harmonization processes and have begun building automated tooling for federated harmonization across silos using synthetic data that will help massively accelerate collaborations at scale (presented at the 2025 NIA Annual Retreat). The availability of methods to produce and analyze massive-scale data has revolutionized scientific investigation. The nature of these data means that they can be used to answer myriad questions and can be mined and augmented on an ongoing basis. Currently, these data primarily exist across clinical, genomic, and genetic domains. However, there is exponential growth in imaging, biologics, and genomics data associated with human samples. Importantly, there is also a recognized need to incorporate more subject-distal data, such as molecular and imaging data from cells derived from patients. To realize the full potential of this data, it is critical to develop methods of data access that facilitate democratization, efficiency, and multi-modal, multi-domain analysis. There is a need for a unified access point that minimizes both administrative and scientific barriers to access while maintaining data integrity and privacy. To approach this, we have focused on both building relationships and building tools. This stems from work in: (1) connecting people to knowledge with tools like OmicSynth, CRISPRBrain, CRISPRLipid, or CARD.AI, (2) connecting people to tools and empowering them by streamlining data harmonization and quality control like GenoTools, FAIRkit, scMAVERICS/scVIRGILS, and ProtPipe, or (3) simply connecting with other investigators and sharing expertise. Milestones and deliverables Completed Deliverables: Data Science Tools and Pipelines: Multi-cloud deployment of GenoML (Status: Completed) scVIRGILS pipeline made publicly available (Status: Completed) scMAVERICS pipeline made publicly available (Status: Completed) ProtPipe V1 made publicly available (Status: Completed) GenoML categorical module made publicly available (Status: Completed) AD/ADRD AI-powered resource catalog alpha release (Status: Completed) AD/ADRD AI-powered resource catalog beta release (Status: Completed) CARD CDEs shared publicly (Status: Completed) Research Outputs: Over 110 publications on PubMed since 2023 attributable to this scope of work Machine learning models built and shared for mitochondrial dysfunction analysis from BLSA and GNPC datasets Machine learning models built and shared for AD/ADRD progression analysis from BLSA and GNPC datasets GenoML platform accessed by over 6.9K unique users in the past 18 months Ongoing Development: scSPATIAL pipeline for spatial transcriptomics (In Progress) ProtPipe V2 enhancement (In Progress) GenoML federated module development (In Progress) Cellular imaging pipeline development (In Progress) Harmonized VUS prediction manuscript (In Progress) AD/ADRD AI-powered resource catalog stable release (In Progress) ML models built and shared for mitochondrial dysfunction from BLSA and GNPC (In Progress) ML models built and shared for AD/ADRD progression from BLSA and GNPC (In Progress) AD/ADRD AI-powered resource catalog stable release (In Progress) CARD CDEs integrated into NLM CDE catalog (In Progress) Major Milestones: AAIC Toronto workshop scheduled for July 2025 NIA Annual Retreat presentation in July 2025 Nature Dementia invited publication targeted for Q4 2025/Q1 2026 References Data Resources: National Library of Medicine's Common Data Elements (CDEs): https://cde.nlm.nih.gov/home Terra.bio cloud platform integration Alzheimer's Disease Data Workbench collaboration UK Biobank harmonization project Finnish Biobank collaboration AllOfUs Study integration Collaborative Networks: University of Mississippi Medical Center partnership UK Dementias Research Institute collaboration SAIL Biobank partnership European Alzheimer's Disease Biobank (EADB) collaboration Global Neurodegenerative Proteomics Consortium (GNPC) Baltimore Longitudinal Study of Aging integration Health Aging and Body Composition Study collaboration Conferences and Workshops: AAIC Workshop presentation scheduled for July 2025 NIA Annual Retreat presentation in July 2025 Board of Scientific Counselors "exceptional" review recognition (2024) Biobank Mining for Modifiable ADRD Risk Factors Executive Summary The Biobank Mining for Modifiable ADRD Risk Factors project represents a comprehensive data science initiative focused on identifying and validating environmental, behavioral, and medical risk factors for Alzheimer's disease and related dementias (ADRD). The project leverages advanced data harmonization techniques and machine learning approaches to analyze large-scale biobank data, including the UK Biobank, Finnish Biobank, AllOfUs Study, and SAIL Biobank. Through systematic analysis of electronic medical records, the team has successfully identified novel associations between viral exposures and neurodegeneration, sleep disturbances, microbiome factors, and various medications that may serve as either risk or protective factors for ADRD development. The project's innovative approach combines standardized Common Data Elements (CDEs) with FAIRkit tooling to accelerate collaborative research across multiple international biobanks. This harmonization strategy has enabled the team to conduct large-scale replication studies and deliver research findings in a streamlined, factory-like manner. The project maintains strong collaborations with the UK Dementias Research Institute, SAIL Biobank, and various internal NIA teams, ensuring alignment with best practices and facilitating knowledge transfer across the research community. Project Summary Narrative Common Data Elements (CDEs) curated using FAIRkit tooling relating to exposures and medications in electronic medical records have allowed the team to accelerate the research of collaborators at the UK Dementias Research Institute as well as the SAIL Biobank and other similar data stewards to facilitate early work on semi-automated metadata harmonization across global repositories. The goal of CARD's Advanced Analytics team is not to reinvent the wheel and build a new data sharing and analysis platform but leverage current gold standard tools after an internal systematic review of similar public offerings. The team has been liaising with internal NIA as well as external/extramural teams to ensure they are following best practices and receive feedback on what they can improve, including the Longitudinal Studies Section and ODSS. As part of this data harmonization strategy, the team facilitates biobank scale collaborations by standardizing electronic medical record codes for both the UK Biobank, Finnish Biobank and AllOfUs Study with special attention paid to AD/ADRD relevant data. They are currently beginning collaborations to accomplish similar harmonization and analysis efforts with the Welsh Biobank in Cardiff (SAIL) that include the development and deployment of standardized code for replication/validation analyses of their findings. One early deliverable of these efforts was an analysis of viral exposures associated with risk of neurodegeneration up to 15 years prior to disease manifestation. The team identified and replicated twenty-two novel pairs of viruses and neurodegenerative diseases in over 500,000 biobank samples as well as replicated the previous association between Epstein-Barr exposure and multiple sclerosis published recently in Neuron. The follow-up to this report includes an in-depth analysis of sleep disturbances as a major contributor to risk of neurodegeneration that has also been published as well as microbiome studies of risk (accepted Science Advances), brain atrophy from MRI as a predictor of AD/ADRD risk and progression, medication exposures as both risk and protective factors as well as genetic interactions using an expanded version of this data and codebase (under review Nature Neuroscience). Upcoming work in this space underway includes hormone treatments in AD/ADRD risk, healthy behaviors such as exercise and diet as protective factors, multi-feature machine learning to predict AD/ADRD risk prodromally as well as a proteomic shift in focus integrating data from the Global Neurodegenerative Proteomics Consortium and UK Biobank to complement the mass spectrometry work being done at CARD. The harmonization and analysis automation has allowed the team to deliver rapidly in an almost factory-like manner. They are also running pilot analyses to evaluate the interactions of genetic risk factors at scale and how these relate to AD/ADRD risk. Milestones and deliverables Completed Deliverables: Sleep analysis results (completed) Gut microbiome analysis results (completed) Medication exposure results (completed) MRI atrophy prediction results (completed) Multi-feature ML for AD/ADRD model (completed) Pilot genetic risk factor interactions in ADRD results (completed) Healthy behaviors, exercise and diet pilot results (completed) Hormone therapy exposure pilot results (completed) In Progress Milestones: Replicate healthy behaviors associations (in progress, on track) Replicate hormone exposure associations (in progress, on track) Replicate genetic interactors in AD/ADRD associations (in progress, on track) Validate multi-feature ML model from electronic medical records (in progress, on track) Validate mitochondrial dysfunction ML model (in progress, on track) Evaluate proteomics outcome models via ML in GNPC and UKB (in progress, on track) Healthy behaviors pre-print draft (Q1 2026) Hormone exposure pre-print (Q1 2026) Genetic interactors pre-print (Q1 2026) Multi-feature ML from EMR pre-print (Q4 2025) Mitochondrial dysfunction ML pre-print (Q2 2026) Proteomics outcome ML pre-print (Q2 2026) Key Achievements: Successfully identified and replicated 22 novel pairs of viruses and neurodegenerative diseases in over 500,000 biobank samples Replicated the association between Epstein-Barr exposure and multiple sclerosis Published findings on sleep disturbances as contributors to neurodegeneration risk Achieved acceptance of microbiome studies in Science Advances Developed automated harmonization and analysis pipeline enabling rapid, factory-like delivery of research findings References Data Sources: UK Biobank (>500,000 samples analyzed) Finnish Biobank AllOfUs Study SAIL Biobank (Welsh Biobank, Cardiff) Global Neurodegenerative Proteomics Consortium Collaborative Networks: UK Dementias Research Institute SAIL Biobank collaboration NIA Longitudinal Studies Section Office of Data Science Strategy (ODSS) NIA/LNG team collaborations Technical Resources: FAIRkit tooling for CDE curation Standardized electronic medical record coding systems Multi-feature machine learning frameworks Automated data harmonization pipelines Cross-platform replication analysis tools Genetic Modifiers of APOE-Mediated Risk for ADRD Executive Summary This project represents the culmination of a three-year international effort to identify genetic modifiers of APOE4 status through comprehensive meta-analysis of harmonized data across global populations. The research has successfully engaged millions of samples from over a dozen sites across Europe and America, establishing one of the largest collaborative networks for APOE4 genetic modifier research. The project has achieved significant milestones including the publication of polygenic risk score findings and submission of genetic risk factor discovery work to Nature Genetics. The research has transitioned from initial discovery phases to advanced multi-omics integration, positioning the project at the forefront of precision medicine applications for Alzheimer's disease and related dementias. Through strategic partnerships with the European Alzheimer's Disease Biobank (EADB) and other international collaborators, the project continues to expand its scope and impact in the field of neurodegenerative disease research. Project Summary Narrative This project is the culmination of a three-year effort to meta-analyze harmonized data across the globe for identification of genetic modifiers of APOE4 status at a genome-wide scale. Currently this work has involved over a million samples and over a dozen sites across Europe and America. We have completed initial tests for polygenic risk, resulting in a manuscript that has been published, as well as genetic risk factor discovery work that is in pre-print now and under review at Nature Genetics. Follow-on portions of this research include multi-omic integrations and collaborations with other NIA teammates as well as collaborators from EADB. Milestones and deliverables Completed Milestones ✅ Build collaborative network for APOE4 research (Complete) ✅ Harmonize data and distribute analysis tools (Complete) ✅ Write manuscript 1 (Complete) ✅ Submit publications and code (Complete) ✅ Harmonized polygenic risk scores (Complete) ✅ Published in Nature Genetics (Complete, see below) Ongoing/In Progress 🔄 Multi-omics integration (In progress) 🔄 Multi-omics integration with EADB for precision medicine applications 🔄 Multi-omics integration with BLSA for precision medicine applications Pending Deliverables ⏳ Summary statistics publicly shared (Pending) ⏳ Manuscript 2 (In progress) Meeting Schedule Monthly meetings with EADB consortium Monthly meetings with Keenan and BLSA team for ongoing collaborations References Published Work Transferability of European-derived Alzheimer's disease polygenic risk scores across multiancestry populations (Nature Genetics) https://www.nature.com/articles/s41588-025-02227-w APOE stratified genome-wide association studies provide novel insights into the genetic etiology of Alzheimers's disease (MedRxiv) https://www.medrxiv.org/content/10.1101/2025.05.07.25327065v1 Collaborative Partnerships European Alzheimer's Disease Biobank (EADB) Baltimore Longitudinal Study of Aging (BLSA) International research sites across Europe and America (12+ institutions) Data Resources Harmonized genetic data from millions of samples Multi-omics datasets for integration analyses Polygenic risk score tools and methodologies Planned Publications Manuscript 2 (in preparation) Multi-omics integration findings (planned) Code and Tools Analysis pipelines for APOE4 genetic modifier identification Harmonized polygenic risk score calculation tools Data harmonization and distribution platforms Data Commons for Multi-omic CRISPR Screens Executive Summary The Data Commons for Multi-omic CRISPR Screens represents a groundbreaking integrated functional genomics platform that leverages CRISPR-based perturbations to systematically map gene function across diverse biological systems. This comprehensive initiative encompasses three specialized platforms: CRISPRbrain, CRISPRlipid, and CRISPRvirus, each targeting critical areas of human health and disease. The project has achieved significant milestones in platform development and data integration, with all three primary platforms successfully launched and operational. The project has demonstrated exceptional impact through its integration with OpenTargets, one of the world's most accessed biomedical knowledge transfer resources. The incorporation of the CRISPR Knowledge transfer tools API endpoint into the early 2025 OpenTargets update represents a major acknowledgment of the program's success and scientific value. This achievement underscores the platform's role in advancing therapeutic development across neurology, metabolism, and infectious disease research. Project Summary Narrative CRISPRbrain, CRISPRlipid, and CRISPRvirus are integrated functional genomics platforms that leverage CRISPR-based perturbations to systematically map gene function across diverse biological systems relevant to human health and disease. CRISPRbrain focuses on uncovering the genetic drivers of brain cell function and neurodegeneration by applying CRISPR screening in iPSC-derived neurons and glia, enabling discovery of therapeutic targets for disorders like Alzheimer's and Parkinson's disease. CRISPRlipid explores lipid metabolism by targeting genes in hepatocytes, adipocytes, and macrophages to reveal regulators of lipid homeostasis implicated in cardiometabolic diseases such as atherosclerosis and fatty liver disease. CRISPRvirus applies genome-scale CRISPR screening to identify host factors that mediate viral infection, immunity, and pathogenesis, accelerating antiviral discovery and pandemic preparedness. Together, these platforms generate rich, cell-type-specific datasets that advance our understanding of complex biological processes and support therapeutic development across neurology, metabolism, and infectious disease. Current stages of maintenance and ongoing development include new features to aid in knowledge transfer as well as ingestion of additional data from the Qi and Ryan teams at CARD. The API endpoint for our CRISPR Knowledge transfer tools has been incorporated into the early 2025 update to OpenTargets, one of the most accessed biomedical knowledge transfer resources in the world. We see this as a major acknowledgment of the success of this program. Milestones and deliverables Completed Deliverables: CRISPRvirus.org platform launched and included basic gene hit search (January 2021) CRISPRlipid.org platform launched with lipid-associated gene browser (January 2022) Cross-platform web framework implementation completed across all CRISPR sites (July 2022) Single-cell result viewer integrated into CRISPRbrain portal (January 2023) Unified CRISPR suite frontend redesign initiated for improved user experience (June 2024) Key Achievements: Successful integration of API endpoint into OpenTargets early 2025 update Establishment of data pipelines for ongoing ingestion of additional data from Qi and Ryan teams at CARD Development of comprehensive knowledge transfer tools across all three platforms Implementation of cell-type-specific dataset generation capabilities Ongoing Initiatives: Continued platform maintenance and feature development Enhancement of user interface and experience across all platforms Expansion of data integration capabilities Development of advanced analytical tools for multi-omic data interpretation References Data Resources: CRISPRbrain.org - CRISPR perturbation and transcriptomic data from UCSF CRISPRlipid.org - CRISPR perturbation and transcriptomic data from UC Berkeley CRISPRvirus.org - CRISPR perturbation and transcriptomic data from UC Berkeley OpenTargets integration - Early 2025 update incorporating CRISPR Knowledge transfer tools API Collaborative Partnerships: University of California, San Francisco (UCSF) - CRISPRbrain data generation University of California, Berkeley - CRISPRlipid and CRISPRvirus data generation CARD Qi and Ryan teams - Ongoing data contribution and collaboration OpenTargets consortium - Knowledge transfer platform integration Technical Infrastructure: DataTecnica platform development and maintenance Multi-platform web framework for unified user experience API endpoints for external platform integration Single-cell data visualization and analysis tools Longitudinal Imaging Harmonization and Analysis via Deep Learning Executive Summary The Longitudinal Imaging Harmonization and Analysis via Deep Learning project represents a cutting-edge initiative to address critical challenges in medical imaging research through advanced artificial intelligence methodologies. This project focuses on developing robust deep learning solutions to harmonize and analyze longitudinal medical imaging data across multiple time points, scanners, and study cohorts, addressing the significant variability in image acquisition protocols that currently limits comparative and pooled longitudinal analyses. The project has demonstrated substantial progress with successful prototype development and validation across multiple imaging datasets including ADNI, PPMI, and UK Biobank. Key achievements include the development of convolutional neural networks with temporal consistency modules, resulting in improved longitudinal accuracy and cross-scanner harmonization capabilities. The project is strategically positioned to advance into proteomic brain aging analysis at scale, working in collaboration with CARD's Advanced Analytics team to focus on clinical outcomes and disease progression modeling. Project Summary Narrative This project aims to develop advanced deep learning methods to harmonize and analyze longitudinal medical imaging data across time points, scanners, and study cohorts. Variability in image acquisition protocols often limits the ability to compare or pool longitudinal data, hindering research into disease progression and treatment response. To address this, we are building robust neural networks that correct for scanner-related and protocol-induced variability while preserving biologically meaningful changes over time. The project integrates domain adaptation, spatiotemporal modeling, and contrast-preserving harmonization techniques, enabling consistent image interpretation across visits and populations. Applied initially to neuroimaging data (e.g., MRI, PET), this approach supports precise tracking of brain aging, neurodegeneration, and treatment effects. The resulting models and tools are designed for scalability and generalization, with potential impact across multiple imaging modalities and disease domains. The next phase of this work is leveraging our work on harmonizing ADNI, PPMI and UK Biobank imaging data to build models of proteomic brain aging at scale in concert with other members of CARD's Advanced Analytics team, focusing on clinical outcomes. Milestones and deliverables The project has achieved significant milestones with several key deliverables completed or in progress: Completed Milestones: Platform development: Web platform successfully developed. Prototype CNN model: Convolutional neural network model trained on multi-site MRI data with successful initial validation (August 2021) Temporal consistency integration: Temporal consistency module successfully integrated, resulting in improved longitudinal accuracy (April 2023) Final model benchmarked and manuscript submitted (February 2025) In Progress Deliverables: Brain aging publication (Under review): Manuscript development focusing on brain aging analysis using harmonized imaging data Research publication on MRI to PET conversion methodologies References Project Management: Project status: On track Google Drive documentation maintained for collaborative access Regular milestone tracking and progress reporting implemented Technical Achievements: Successful prototype CNN model validation across multi-site MRI datasets Implementation of temporal consistency modules for improved longitudinal accuracy Cross-scanner harmonization methodology development and validation Collaborative Frameworks: Integration with CARD's Advanced Analytics team for proteomic brain aging analysis Multi-cohort data harmonization across ADNI, PPMI, and UK Biobank datasets Scalable model development for broad application across imaging modalities and disease domains Publications MRI2PET: Realistic PET Image Synthesis from MRI for Automated Inference of Brain Atrophy and Alzheimer's (under review) https://pubmed.ncbi.nlm.nih.gov/40313301/ Application of Aligned-UMAP to longitudinal biomedical studies (published, Patterns) https://pubmed.ncbi.nlm.nih.gov/37409055/ Multimodal Patient Representation Learning with Missing Modalities and Labels (published, International Conference on Learning Representations (ICLR) 2024) https://openreview.net/forum?id=Je5SHCKpPa Prediction, prognosis and monitoring of neurodegeneration at biobank-scale via machine learning and imaging (preprint) https://pubmed.ncbi.nlm.nih.gov/39574848/ SECONDGRAM: Self-conditioned diffusion with gradient manipulation for longitudinal MRI imputation (published, Patterns) https://pubmed.ncbi.nlm.nih.gov/40486972/ Transferability of European-derived Alzheimer's disease polygenic risk scores across multiancestry populations (Nature Genetics) https://www.nature.com/articles/s41588-025-02227-w APOE stratified genome-wide association studies provide novel insights into the genetic etiology of Alzheimers's disease (MedRxiv) https://www.medrxiv.org/content/10.1101/2025.05.07.25327065v1 Genome-wide CRISPRi/a screens in human neurons link lysosomal failure to ferroptosis (published, Nature Neuroscience) https://pubmed.ncbi.nlm.nih.gov/34031600/ A CRISPRi/a platform in human iPSC-derived microglia uncovers regulators of disease states (published, Nature Neuroscience) https://pubmed.ncbi.nlm.nih.gov/35953545/ Parallel CRISPR-Cas9 screens identify mechanisms of PLIN2 and lipid droplet regulation (published, Developmental Cell) https://pubmed.ncbi.nlm.nih.gov/37494933/ Sleep disturbances as risk factors for neurodegeneration later in life (published) https://www.nature.com/articles/s44400-025-00008-0 Medication Exposure and Neurodegenerative Disease Risk Across National Biobanks (under review, Nature Neuroscience) https://www.medrxiv.org/content/10.1101/2025.07.07.25330740v1 Gut-Brain Nexus: Mapping Multi-Modal Links to Neurodegeneration at Biobank Scale (accepted, Science Advances) https://pubmed.ncbi.nlm.nih.gov/39371139/ MRI2PET: Realistic PET Image Synthesis from MRI for Automated Inference of Brain Atrophy and Alzheimer's (preprint) https://www.medrxiv.org/content/10.1101/2025.04.23.25326302v1.full.pdf GenoTools publication: https://doi.org/10.1093/g3journal/jkae268 Over 110 publications on PubMed since 2023 attributable to data harmonization scope of work Web Resources and Platforms GenoML automated machine learning platform: https://genoml.com/ GenoTools Python package: https://pypi.org/project/the-real-genotools/ scMAVERICS pipeline: https://github.com/NIH-CARD/scMAVERICS scVIRGILS pipeline: https://github.com/NIH-CARD/scVIRGILs Longitudinal GWAS pipeline: https://longitudinal-gwas-pipeline.readthedocs.io/en/latest/ CARD Catalogue beta version: https://cardcatalogue-beta.streamlit.app/ CRISPRbrain.org - CRISPR perturbation and transcriptomic data from UCSF CRISPRlipid.org - CRISPR perturbation and transcriptomic data from UC Berkeley CRISPRvirus.org - CRISPR perturbation and transcriptomic data from UC Berkeley National Library of Medicine's Common Data Elements (CDEs): https://cde.nlm.nih.gov/home Terra.bio cloud platform integration OpenTargets integration - Early 2025 update incorporating CRISPR Knowledge transfer tools API GitHub Repositories https://github.com/NIH-CARD/scMAVERICS https://github.com/NIH-CARD/scVIRGILs