Interpretable Machine Learning Links Group-Specific Plasma Proteomics to Regional Structural Neuroimaging in Alzheimer's Disease

In ADNI participants, plasma proteomic group-specific signatures classified CN vs MCI (AUC 0.956 ± 0.017) and MCI vs AD (AUC 0.963 ± 0.011), with CN vs AD at AUC 0.836 ± 0.085.

• Why it matters: AD pathology precedes symptoms, but it remains unclear how scalable plasma proteomics relates to AD-related regional brain neurodegeneration across the continuum.
• What they did: Using 146 plasma analytes in 565 ADNI participants, pairwise ML classification with nested 5×5 cross-validation and SHAP prioritized proteins, then ordinary least squares regressions tested protein–MRI links across 103 ComBat-harmonized structural volumes with FDR correction.
• The result: No protein–MRI associations survived FDR for CN vs MCI; in MCI vs AD, cortisol and osteopontin were linked to reduced left precuneus/superior parietal and left paracentral/accumbens volumes, and in CN vs AD, vitronectin was linked to lower posterior corpus callosum volume.

AI summary of the abstract · may contain errors · not medical advice · read the paper.