Summary: Diagnosing neurodegenerative diseases is challenging because symptoms often overlap—a single patient can show signs of Alzheimer’s disease, Lewy body pathology, and the effects of a small stroke simultaneously. Researchers have now developed an AI model that can detect multiple neurodegenerative conditions from a single blood sample by recognizing disease-specific protein patterns.
By analyzing plasma proteomics from a large database of more than 17,000 individuals, the model identified distinct biological signatures associated with Alzheimer’s disease, Parkinson’s disease, amyotrophic lateral sclerosis (ALS), frontotemporal dementia, and vascular damage from prior stroke. The study suggests that a protein-based profile can predict cognitive decline more accurately than a solely symptom-based clinical diagnosis.
Key Facts
- Joint learning approach: Instead of training separate models for each disorder, the AI uses a joint-learning strategy to discover common and distinct protein patterns across multiple brain-degenerative conditions.
- Multiple conditions from one blood draw: The model discriminates among Alzheimer’s, Parkinson’s, ALS, frontotemporal dementia and vascular injury related to stroke.
- Biological subtypes revealed: Patients with the same clinical diagnosis often display different underlying protein profiles, indicating that individual biological subtypes exist and that one-size-fits-all treatments may be ineffective.
- Largest proteomics resource: The AI was trained on the Global Neurodegeneration Proteomics Consortium (GNPC) database, the largest proteomic repository for neurodegenerative diseases.
Source: Lund University
Context: Symptoms of neurodegenerative diseases frequently overlap, and age-related cognitive decline can result from several coexisting brain pathologies. That overlap complicates diagnosis and clinical management because current clinical assessments primarily rely on observed symptoms, which are not always specific to a single underlying disease process.
Researchers at Lund University, together with contributors from the Swedish BioFINDER study and GNPC, developed an AI model that uses plasma protein measurements to identify disease-linked patterns and provide probabilistic diagnoses across several dementia-associated conditions.
The model, described in Nature Medicine, was trained on proteomic data from over 17,000 patients and control participants aggregated across multiple GNPC datasets. Using advanced statistical learning and joint-learning strategies, the team extracted a set of proteins that together indicate a general pattern of brain degeneration and then used that pattern to classify specific disorders.
Lead researcher Jacob Vogel and first author Lijun An report that their model outperformed previous approaches and achieved strong diagnostic accuracy across conditions. An important advantage of the method is that its results were validated across multiple independent datasets, strengthening confidence in the model’s generalizability.
The researchers also found that the protein-based classifications correlated with future cognitive decline and revealed subgroups of patients with apparent co-pathologies. In many cases, individuals clinically diagnosed with Alzheimer’s exhibited protein signatures more characteristic of other disorders, suggesting mixed pathology, alternative disease trajectories, or potential misdiagnosis under standard clinical practice.
Vogel cautions that current blood-based protein measurements are not yet sufficient to replace comprehensive clinical assessment. He emphasizes the need to refine the method and combine proteomic testing with other diagnostic tools. Still, many proteins highlighted by the model point to biological pathways that merit follow-up studies and could improve understanding of disease mechanisms.
Next steps for the research team include expanding the proteomic marker panel using more sensitive technologies such as mass spectrometry to further separate disease-specific patterns and improve diagnostic specificity. The long-term goal is a reliable, scalable blood test that supports diagnosis across neurodegenerative disorders without reliance on invasive procedures.
Facts
GNPC
The Global Neurodegeneration Proteomics Consortium (GNPC) is an international collaboration that has assembled one of the world’s largest databases of protein measurements related to neurodegenerative disease. GNPC enables large-scale analysis of proteomic data to accelerate the discovery of biomarkers and better understand disease biology.
Proteomics
Proteomics studies the full set of proteins present in a biological sample. In a blood sample, proteomic analysis quantifies the levels of many proteins simultaneously, revealing patterns that reflect biological processes and disease states. These protein patterns can serve as biomarkers for early detection, prognosis, and treatment stratification.
Key Questions Answered:
A: Clinical diagnosis often depends on symptoms such as memory loss, movement problems or behavioral changes, which can arise from different underlying diseases. Proteomic testing measures molecular signals—proteins released into the blood by brain tissues—that can reveal disease processes missed or misclassified by symptom-based assessment.
A: The researchers observed many cases where individuals clinically diagnosed with Alzheimer’s had protein signatures typical of other disorders, supporting the presence of mixed pathology. The joint-learning model can assign probabilistic scores for multiple conditions, helping to unmask overlapping disease layers.
A: Although this work represents a major research milestone, the method requires further refinement and validation before it becomes a routine clinical test. The authors plan to expand marker panels and validate the approach in additional clinical settings; for now, the model is a powerful research and clinical-trial tool.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal article was reviewed in full by the editorial team.
- Additional context and clarifications were added by staff to improve readability.
About this AI and dementia research news
Author: Anna Elizabeth Hellgren
Source: Lund University
Contact: Anna Elizabeth Hellgren – Lund University
Image credit: Neuroscience News
Original Research: Open access. Title: “A deep joint-learning proteomics model for diagnosis of six conditions associated with dementia” by Lijun An et al., published in Nature Medicine. DOI: 10.1038/s41591-026-04303-y
Abstract
A deep joint-learning proteomics model for diagnosis of six conditions associated with dementia
Co-pathology—multiple overlapping disease processes—is common in neurodegenerative disorders and complicates diagnosis and treatment. Sensitive, specific and scalable in vivo biomarkers for many neuropathologies remain unavailable. The authors present ProtAIDe-Dx, a deep joint-learning model trained on plasma proteomic data from 17,187 participants (mean age 70.3 ± 11.5 years; 53.2% female). ProtAIDe-Dx provides simultaneous probabilistic diagnoses across six conditions associated with dementia and achieves cross-validated balanced classification accuracy between 70–95% and area under the curve greater than 78% across conditions.
The model’s diagnostic probabilities uncovered patient subgroups with co-pathologies and correlated with pathology-specific biomarkers in an independent memory clinic sample, including individuals without cognitive impairment. Interpreting the model highlighted protein networks that mark both shared and disease-specific biological processes and identified proteins known and novel to these conditions. ProtAIDe-Dx improved biomarker-based differential diagnosis at the patient level, indicating proteins that drive individual diagnostic decisions. Overall, the results underscore the potential of plasma proteomics to enhance diagnostic workup from a single blood draw.