‘We built a brain roadmap for each person’: New AI-driven tools help decode Alzheimer’s disease at the cellular level

Alzheimer’s disease is complex, presenting and progressing in so many different ways in individual patients that it’s difficult for scientists to study. Now, researchers at the University of Wisconsin–Madison have developed AI-driven bioinformatics tools to make it easier to investigate the cellular and genetic changes driving the disease and link them to patients’ symptoms, with the potential to improve diagnosis and pinpoint therapeutic targets.

Alzheimer’s is a neurodegenerative disease, causing a slow decline in cognitive function and related effects around the body. It results from a complex web of molecular and cellular changes that influence and overlap with one another. When studying the disease, it can be difficult to determine which biological changes are related to specific symptoms, information that could ultimately be used to design new treatments.

The new tools, designed by researchers in the lab of Daifeng Wang, a UW–Madison professor of biostatistics, medical informatics and computer sciences, and an investigator in UW–Madison’s Waisman Center, aim to help researchers decipher the molecular underpinnings of the condition by linking changes in brain cells to Alzheimer’s symptoms, identifying specific genes and cell populations involved in disease progression and cognitive resilience, and creating gene roadmaps for individuals with the condition.

Called PASCode and iBrainMap, and published recently in Nature Medicine and Nature Communications, respectively, the new tools were motivated and designed using an emerging AI model called graph neural networks and built on a dataset of more than 6.3 million cells from the brains of nearly 1,500 autopsy donors.

PASCode and iBrainMap are part of a new group of Alzheimer’s findings by the PsychAD Consortium, a National Institutes of Health-supported effort led by Wang and Panos Roussos, professor of genetics and genomics sciences at Mount Sinai Medical School. The consortium includes researchers around the country working to identify biomarkers and therapeutic targets for Alzheimer’s disease and neuropsychiatric symptoms.

PASCode helps researchers identify which cellular changes and alterations in the way genes are expressed may be linked to specific Alzheimer’s symptoms, including depression, as well as the ways brains remain resistant to cognitive decline. “The conceptual idea of this work is to transfer the donor-level clinical information, like symptoms, to the single cells. With PASCode, we can score each individual cell from each donor on how likely the cell is associated with particular clinical phenotypes,” Wang says.

PASCode assigns each cell a score that reflects how strongly it is associated with specific Alzheimer’s symptoms. The stronger the score, the more likely the cell is involved in disease-related changes.

After scoring all 6.3 million cells, PASCode found that about 1.5 million were associated with characteristics of Alzheimer’s disease. Researchers then used those data to investigate which groups of cells contribute most to Alzheimer’s disease, which cellular changes are linked to disease progression and cognitive resilience, and how Alzheimer’s disease and depression may be connected.

Wang and his team also compiled the findings into a publicly available atlas that researchers can use to study Alzheimer’s and associated neuropsychiatric symptoms, such as characterizing the disease-implicated cell types from their own data.

Because Alzheimer’s disease varies widely in symptoms and progression, the researchers wanted to examine each individual person separately rather than compare broad groups of healthy and affected individuals. Wang and his team further used the PsychAD dataset in AI-based personalized analysis of cell types and genes, examining how DNA mutations disrupt the way genes work together in each individual with Alzheimer’s disease.

Using iBrainMap, also based on a graph neural network approach, “we built a brain gene roadmap for each person,” Wang says.

This roadmap identifies cell-type specific genes that may influence disease progression, measured through pathology and cognition, and whether those genes are associated with specific symptoms. The new personalized roadmap improved classification of healthy individuals versus individuals with Alzheimer’s disease by 21%.

In particular, the researchers examined three neuropsychiatric symptoms associated with Alzheimer’s disease: weight loss, insomnia, and depression.

“For each of these symptoms, we have an important gene list, important cell types and an important gene network for each donor,” Wang says.

Based on these personalized analyses, they were also able to identify new Alzheimer’s disease subtypes, implying there may be groups of Alzheimer’s patients with shared molecular and cellular characteristics that haven’t been described yet.

“The far reach is that these kinds of analyses should lead to more personalized and targeted therapies,” says Pramod Bharadwaj Chandrashekar, the research scientist in Wang’s lab who led the study. Discovering person-specific genes and cell types may be a first step toward identifying personalized treatment targets.

This research was supported in part by a grant from the National Institutes of Health (R01AG067025).