AI Uncovers New Alzheimer’s Disease Mechanism and Identifies Potential Treatment

A groundbreaking new study has revealed that a gene previously considered a biomarker for Alzheimer’s disease (AD) is, in fact, a key causal factor. Researchers from the University of California, San Diego (UCSD) have used artificial intelligence (AI) to uncover a hidden pathogenic mechanism and identified a potential drug that can inhibit the gene’s “moonlighting” function.

The findings were published on April 23, 2025, in the prestigious journal Cell.

Alzheimer’s disease is the most common cause of dementia. While certain rare genetic mutations are known to cause the disease, they account for only a small fraction of all cases. Most patients develop sporadic Alzheimer’s disease, the causes of which remain largely unknown. Identifying these hidden causes could significantly improve disease management and treatment.

Prof. Sheng Zhong, the lead researcher and professor at the Shu Chien-Gene Engineering Department of Bioengineering at UCSD, said:
“Currently, treatment options for Alzheimer’s disease are very limited and largely ineffective.”

The research team focused on the gene PHGDH (phosphoglycerate dehydrogenase). They had previously identified PHGDH as a blood biomarker for early-stage AD. Later studies showed that the expression level of PHGDH correlated directly with brain damage severity — the higher the expression, the more severe the disease. This correlation was validated across several patient cohorts from different medical centers.

To determine whether PHGDH plays a causal role in the disease, researchers used mouse models and human brain organoids. They found that reducing PHGDH expression slowed disease progression, while increasing it worsened symptoms. These results confirmed that PHGDH is indeed a pathogenic gene responsible for sporadic Alzheimer’s disease.

Using AI, the team further discovered that the PHGDH protein has an unexpected, previously unknown function: it interferes with gene regulation in brain cells, triggering neurodegenerative disorders like Alzheimer’s disease.

The Gene’s “Moonlighting” Function

Until now, PHGDH was known only as a critical metabolic enzyme involved in the synthesis of serine — an essential amino acid and neurotransmitter. It was once speculated that its metabolic activity might be linked to AD, but this hypothesis remained unproven.

Prof. Zhong said,
“Our research hit a bottleneck — we couldn’t figure out the exact mechanism at work.”

A breakthrough came when the team’s earlier research showed widespread gene regulatory imbalances in the brains of Alzheimer’s patients. This led them to suspect that PHGDH might have an additional, unknown regulatory function.

Turning once again to AI, they predicted the protein’s 3D structure and discovered a region resembling the DNA-binding domain of a transcription factor. Although this structure was only spatially similar — not sequence-related — it opened new doors.

“This discovery would not have been possible without the precision of AI predictions,” Zhong said.

The researchers confirmed that this hidden structure could activate two critical genes, disrupting normal gene regulation in the brain and ultimately leading to AD. This revealed that PHGDH possesses a previously unrecognized regulatory function — a key contributor to sporadic Alzheimer’s disease.

The study also reinforced the link between PHGDH expression levels and disease severity. While everyone carries the PHGDH gene, it’s the level of gene expression, i.e., the amount of protein produced, that determines disease risk.

A New Window for Treatment

With a clearer understanding of the pathogenic mechanism, the team turned its attention to potential therapeutic interventions.

Current AD treatments largely focus on removing beta-amyloid plaques in the brain, but by the time these plaques form, it is often too late. The newly identified gene regulation pathway appears earlier in the disease process, making it a more promising intervention point.

Using AI-driven drug screening, the team identified a small molecule inhibitor called NCT-503. This molecule does not significantly disrupt PHGDH’s metabolic activity (serine synthesis) and can cross the blood-brain barrier. AI simulations showed that NCT-503 precisely binds to the DNA-binding region of PHGDH, effectively blocking its regulatory function.

In two mouse models of Alzheimer’s disease, NCT-503 significantly improved cognitive performance and reduced anxiety behaviors. While limitations remain — including the lack of perfect animal models for sporadic AD — the results are encouraging.

“We now have a validated drug candidate with therapeutic potential,” said Zhong. “This opens the door for clinical development and introduces a new class of small-molecule treatment strategies.”

Unlike current infusion therapies, small-molecule drugs like NCT-503 could be administered orally, offering greater convenience.

The next step is to further optimize the compound and prepare for preclinical studies with the U.S. Food and Drug Administration (FDA).

Paper Link:

“Transcriptional Regulatory Activity of PHGDH Drives Amyloid Pathology in Alzheimer’s Disease”
Link: https://www.cell.com/cell/fulltext/S0092-8674(25)00397-6