Brain Region-Specific Peptide and Proteomic Signatures Inform Alzheimer’s Disease Biomarker Panels

Abstract Alzheimer’s disease (AD) is a heterogeneous neurodegenerative disorder with region- and severity-specific molecular signatures that complicate biomarker discovery. We combined peptide-centric quantitative proteomics data with machine learning, dimensionality reduction, nonlinear modeling, and network enrichment to identify candidate biomarkers across four brain regions and four AD severity stages. Dimensional reduction revealed disease-severity gradients (HCF-low ADNC → HCF-high ADNC → sporadic AD dementia (ADD) → autosomal-dominant AD) that were strongest in cortical regions (SMTG, IPL), moderate in the hippocampus, and weakest in the caudate. Machine-learning-based feature selection improved severity prediction, particularly at the peptide level, yielding panels of 10–22 region- and severity-specific biomarkers. Biomarker interaction modeling identified hubs centered on haptoglobin, VGF, APP, APOE, and midkine family members, partnering within neurosecretory, lipid transport, acute-phase response, and proteostasis pathways. STRING analysis validated these interactions, showing enrichment for secreted, hemoglobin-binding proteins with plasma and bone marrow tissue signals. Although biomarkers were predominantly region-specific, they converged on blood-linked pathways, supporting the translational potential of multi-marker, circulation-accessible panels. These findings show that peptide-level proteomics integrated with machine learning and network analysis clarifies AD heterogeneity and points to therapeutic avenues in neurovascular integrity, lipid transport, acute-phase response, and proteostasis, accelerating AD biomarker development.