Researchers are using artificial intelligence to move beyond one-size-fits-all Alzheimer's diagnosis, identifying distinct disease subtypes that could unlock personalized treatments. A major new initiative called AI4AD2, backed by a $12.6 million NIH grant that brings total investment to $30.7 million, aims to decode the biological complexity of dementia by analyzing brain imaging, genetic sequences, and other health data from tens of thousands of patients.

Why Does Alzheimer's Look Different in Every Patient?

The challenge with treating Alzheimer's disease is that no two brains degenerate in exactly the same way. Some patients develop amyloid plaques, others accumulate tau tangles, and many experience vascular damage or Parkinson's-like changes simultaneously. These different pathological processes unfold at different speeds in different people, making it nearly impossible to design treatments that work broadly across all patients.

"As we age, our brains decline. But each of us has a unique mix of degenerative processes going on in our brains. We may have a mix of Alzheimer's pathology, vascular disease, and brain changes more typical of Parkinson's disease, all of them proceeding at different rates. This mix of pathologies makes dementia hard to treat," said Paul M. Thompson, associate director of the USC Mark and Mary Stevens Neuroimaging and Informatics Institute.

Paul M. Thompson, Associate Director, USC Mark and Mary Stevens Neuroimaging and Informatics Institute

AI4AD2, led by researchers at USC and involving 10 investigators and 23 co-investigators from 10 institutions, is designed to solve this problem by using machine learning to identify meaningful subtypes of Alzheimer's and related dementias. Instead of grouping all patients under one diagnostic label, the AI system will categorize individuals based on patterns in their brain scans, cognitive test results, tissue pathology, and genetic data.

How Will AI Identify These Disease Subtypes?

The project will employ several cutting-edge AI techniques to uncover hidden disease patterns. One major innovation involves developing "genomic language models," a type of artificial intelligence inspired by the same technology that powers ChatGPT and other large language models (LLMs). Instead of analyzing words, these models will analyze DNA sequences to identify combinations of genetic changes associated with Alzheimer's disease, disease progression, and key biological markers.

The AI4AD2 team will train and evaluate these methods using data from over 58,000 participants across 57 different research cohorts. This represents an unprecedented scale for Alzheimer's research. The goal is to teach the AI to search vast genetic datasets for patterns that traditional statistical methods cannot detect, uncovering new genetic and protein-related changes that may drive neurodegeneration and linking them to measurable changes in the brain and behavior.

Earlier AI4AD research demonstrated the potential of this approach. AI models trained on 80,000 brain scans could identify Alzheimer's-related features with over 90% accuracy, showcasing what becomes possible when combining imaging, genomics, and machine learning at scale.

Steps to Translate AI Discoveries Into Better Patient Care

• Subtype-Specific Clinical Trials: By identifying distinct disease subtypes, researchers can design clinical trials that match patients with treatments most likely to benefit their particular biological profile, improving trial success rates and reducing time to approval.
• Genome-Guided Drug Discovery: Using a tool called PreSiBO, an AI-based drug discovery system developed through the original AI4AD effort, researchers will identify subtype-specific therapeutic targets and determine whether existing drugs can be repurposed for patients with specific Alzheimer's-related biological profiles.
• Personalized Medicine Development: By understanding which molecular pathways are affected in each patient subtype, researchers can develop AI tools to identify specific drug treatments that target these distinct disease mechanisms, moving toward truly personalized Alzheimer's care.

What About Patients From Different Ancestry Groups?

A critical limitation of current biomedical research is that many datasets focus heavily on people of European ancestry, which limits the ability to identify risk factors that may affect other populations differently. AI4AD2 is explicitly designed to address this disparity by adapting its disease classification, subtyping, and prognosis tools for global and multi-ancestry cohorts.

The project will incorporate datasets from African, Indian, Korean, and US populations, and will identify how ancestry, social, and environmental factors affect Alzheimer's risk and progression. This approach aims to develop more accurate predictive models that work across diverse populations, ensuring that AI-driven insights benefit patients worldwide rather than just those represented in existing research databases.

"Artificial intelligence is only as powerful as the data and scientific questions behind it. This renewal allows our team and collaborators to work at a scale that was previously out of reach, integrating imaging, genomics, and other biomarkers to better capture the complexity of Alzheimer's disease. It represents an important step toward more precise, inclusive, and actionable brain health research," said Arthur W. Toga, director of the USC Mark and Mary Stevens Neuroimaging and Informatics Institute.

Arthur W. Toga, Director, USC Mark and Mary Stevens Neuroimaging and Informatics Institute

What Makes This Different From Previous Alzheimer's Research?

The original AI4AD initiative, launched in 2020, developed AI tools to detect Alzheimer's-related patterns in brain scans and showed how machine learning could link imaging findings to underlying genetic risk. AI4AD2 builds on this foundation but goes much further by integrating multiple data types simultaneously and focusing on disease subtyping and precision treatment discovery.

The project represents a fundamental shift in how researchers approach dementia. Rather than asking "Does this patient have Alzheimer's disease?" the new initiative asks "What specific combination of biological processes is driving this patient's cognitive decline, and what treatment would work best for their particular disease profile?" This distinction matters enormously for drug development and clinical practice.

The Stevens INI will serve as the major hub for the effort, with partner institutions contributing expertise in neuroimaging, genomics, statistics, machine learning, cognitive science, and drug discovery. Importantly, the team plans to share software and tools via public repositories and scientific workshops so that researchers worldwide can use and build upon the project's methods, accelerating progress across the entire field.

For families affected by Alzheimer's disease, the long-term goal is clear: to develop more accurate tools to distinguish different types of dementia and identify the best therapies for individual patients. By combining large-scale data with advanced AI, AI4AD2 seeks to bring personalized medicine closer to reality for one of the world's most devastating neurological diseases.