Interpretable Machine Learning Links Group-Specific Plasma Proteomics to Regional Structural Neuroimaging in Alzheimer’s Disease
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Alzheimer’s disease (AD) pathology begins years before clinical symptoms, highlighting the need for scalable biomarkers that capture biological changes across the AD continuum. Broader plasma proteomics may capture systemic processes that vary with disease stage, yet their relationship to AD-related brain neurodegeneration remains unclear. We aimed to identify plasma proteomic signatures across cognitively normal (CN), mild cognitive impairment (MCI), and AD and evaluate their neuroanatomical associations.
Methods
We analyzed 146 plasma analytes from 565 Alzheimer’s Disease Neuroimaging Initiative (ADNI) participants (58 CN, 396 MCI, 111 AD). Group-specific proteomic signatures were identified using pairwise classification models with nested 5×5 cross-validation. SHAP quantified feature importance, with the top 15 proteins per comparison being prioritized for MRI association analysis. We analyzed 103 regional structural MRI volumes harmonized across 57 acquisition sites using ComBat. Separate ordinary least squares regression tested protein-MRI associations, adjusting for age, sex, APOE ε4, diagnosis, and intracranial volume, followed by Benjamini-Hochberg false discovery rate (FDR) correction.
Results
Classification models using proteomic data achieved AUC of 0.956 ± 0.017 for CN vs MCI, 0.963 ± 0.011 for MCI vs. AD, and 0.836 ± 0.085 for CN vs. AD. SHAP identified group-dependent proteomic signatures. No protein-MRI associations survived FDR correction for CN vs. MCI. In MCI vs. AD, higher cortisol was associated with lower left precuneus and superior parietal volumes, while higher osteopontin was associated with lower left paracentral and accumbens volumes. In CN vs. AD, higher vitronectin was associated with lower posterior corpus callosum volume.
Conclusion
These findings suggest that plasma proteomics capture group-specific systemic biological signals across the AD continuum. Linking ML-prioritized plasma proteins to regional brain structure provides neuroanatomical context for blood-based biomarker discovery.
Article activity feed
-
Interpretable Machine Learning Links Group-Specific Plasma Proteomics to Regional Structural Neuroimaging in Alzheimer's Disease #NeuroDegeneration 🧪🧠 https://www.biorxiv.org/content/10.64898/2026.11.02.763368v1
-
Interpretable Machine Learning Links Group-Specific Plasma Proteomics to Regional Structural Neuroimaging in Alzheimer's Disease https://www.biorxiv.org/content/10.64898/2026.11.02.763368v1
-
Interpretable Machine Learning Links Group-Specific Plasma Proteomics to Regional Structural Neuroimaging in Alzheimer's Disease https://www.biorxiv.org/content/10.64898/2026.11.02.763368v1
-
