AI-Powered Project Seeks to Predict Protein Dysfunction in Neurodegenerative Diseases

Alzheimer's, Parkinson's, ALS, and Huntington's disease all share a common signature: proteins that lose their shape, clump together, and destroy cells. By the time a doctor can see these changes clearly on a scan or in a lab test, the damage is often already extensive. A new research initiative, backed by $28.6 million in federal funding, aims to catch that process much earlier—before the cascade begins—by training artificial intelligence to predict which proteins are about to malfunction.

The challenge is that roughly one-third of all proteins in the human body don't have a fixed structure. These intrinsically disordered proteins, or IDPs, shift constantly among different shapes, and that flexibility is precisely what makes them dangerous. They can tangle with other proteins in ways that trigger neurodegeneration. Current AI systems excel at predicting the behavior of stable proteins, but IDPs remain largely opaque. "Treatments for neurodegenerative diseases remain extremely limited," said Randal Halfmann, an investigator at the Stowers Institute for Medical Research in Kansas City. "Decoding the language of these IDP interactions could allow us to create therapeutic IDPs to 'intercept' problematic interactions that drive diseases like Alzheimer's."

Halfmann's laboratory will anchor the new project, called NATIVE-ID, by generating the massive dataset that AI systems need to learn. His team will receive approximately $4.1 million over two years from the Advanced Research Projects Agency for Health, or ARPA-H. The work will unfold across more than one million samples, producing over 10 billion individual measurements of how proteins aggregate. To accomplish this, Halfmann will test 50,000 different proteins, each expressed separately in yeast cells under conditions designed to replicate what happens inside human cells as we age. He will use a technology his own lab developed in 2018 called Distributed Amphifluoric FRET, or DAmFRET, which measures protein self-assembly inside living cells with cellular precision. "Direct measurements of protein aggregation at cellular resolution have not previously been done at this scale," Halfmann said.

The NATIVE-ID project brings together scientists from UC Berkeley, Johns Hopkins University, Brown University, Emory University, Texas A&M University, and other institutions. UC Berkeley's Innovative Genomics Institute leads the effort. Each partner contributes a distinct capability: some will generate human neurons from different genetic backgrounds, others will measure protein interactions in those neurons, still others will develop the deep-learning frameworks needed to interpret the data and run large-scale computational simulations. Alejandro Sánchez Alvarado, president and chief scientific officer at the Stowers Institute, emphasized that neither component of the collaboration could succeed alone. "A deep learning model that is to decode the language of disordered proteins needs rigorously obtained data at scale, something that has yet to exist," he said. "Randal's laboratory will provide what has been missing: more than 10 billion measurements of aggregation, taken one living cell at a time, across 50,000 proteins. Our colleagues at the Innovative Genomics Institute will bring the human neurons in which the model's predictions must ultimately hold."

The team will initially focus on frontotemporal lobar degeneration, or FTLD, a neurodegenerative disease that shares genetic and biological features with ALS. The work builds on Halfmann's prior research: in 2023, his lab became the first to experimentally determine the structure of the initiating step in amyloid formation associated with Huntington's disease. His team has also studied TDP-43, a protein strongly implicated in both ALS and FTLD. Those earlier studies examined hundreds of protein sequences. This new effort will scale that work to 50,000. "Those were sort of pilots for this new undertaking," Halfmann said.

The potential payoff extends beyond basic science. If researchers can predict when proteins are likely to misfold and at what age disease onset might occur, patients could seek preventive treatments or enroll in clinical trials before significant neurological damage accumulates. "If we can better predict the probabilities and onset ages of disease, it could allow many more people to seek preventive or early-stage treatments or enroll in clinical trials," Halfmann said. The current award funds an initial two-year phase focused on generating foundational datasets and models. A second phase, also planned for 24 months, will focus on validating potential therapeutics and developing methods to detect protein dysfunction earlier in the disease process. The work is part of BIOGAMI, a broader ARPA-H initiative aimed at understanding and ultimately controlling harmful protein aggregation across multiple neurodegenerative conditions.