AI and Neurological Rare & Neuroimmune Diseases ── Parkinson's ── 2026-11-21

Parkinson's disease and parkinsonism: treatment, trials and AI
- Abstract
Parkinson’s disease (PD) progresses heterogeneously, yet no reliable biomarker stratifies patients for individualized care. We developed DEGAS, a machine learning framework applying functional connectivity to identify network-defined PD subtypes. Baseline functional MRI from 336 early-stage patients was stratified by suspected REM sleep behavior disorder (sRBD), a marker of aggressive disease. Adjacency matrices and graph theory metrics were integrated into classifiers, and predictions validated against three-year outcomes. Region-of-interest analyses showed no group differences in connectivity or topology. By contrast, models incorporating graph theory features achieved robust classification (balanced accuracy 0.91–0.93; MCC 0.61–0.78), outperforming adjacency-only models (balanced accuracy 0.79; MCC 0.52). Metrics that represented the basal ganglia-thalamic, temporal-limbic, and cerebello-cortical circuits were the strongest predictors. Patients classified as sRBD-positive showed worse motor outcomes, accelerated autonomic dysfunction, and higher rates of hallucinations, dyskinesias, and depression. The DEGAS framework defines a baseline network signature marking clinically distinct PD subgroups. External validation is required before clinical use is considered.
Journal IF-equivalent: 8.0 (OpenAlex 2-year mean citedness, value as of 2026-10-20, retrieved 2026-10-21; not the official Clarivate IF)Reference: Martinez-Nunez AE, Pontone GM, Barmpoutis A, Okun MS, Vaillancourt DE, Wong JK. DEGAS, disease endophenotyping through graph and adjacency metrics: network biomarkers for Parkinson’s disease progression. npj Parkinsons Dis. 2026 Nov 18 [Epub ahead of print]. doi:10.1038/s41531-026-01602-6.Checked: Abstract only - Abstract
Parkinson’s disease (PD) is a progressive neurodegenerative disorder, yet reliable and minimally invasive biomarkers for accurate diagnosis and clinical stratification remain lacking. Here, we aimed to identify plasma-based protein biomarkers with diagnostic and differential utility for PD. Plasma samples from 391 participants were analyzed using a two-stage design. In the discovery phase, label-free quantitative proteomics identified differentially expressed proteins between patients with PD and healthy controls, followed by functional enrichment analyses. A machine learning–based feature selection strategy was used to derive candidate biomarker panels. In the validation phase, selected proteins were quantified by enzyme-linked immunosorbent assay in an independent cohort including patients with PD, patients with multiple system atrophy, Parkinsonian type (MSA-P), and healthy controls. Diagnostic performance was evaluated using receiver operating characteristic analyses, and associations with clinical features were examined. Proteomic profiling identified 121 differentially expressed proteins related to vesicle trafficking, oxidative stress, and autophagy. A plasma protein panel comprising LAMP1, APOD, VIM, CTSF, and HEG1 showed robust performance in distinguishing PD from healthy controls, while a simplified combination moderately differentiated PD from MSA-P. Selected biomarkers were also associated with clinical scales of cognition and non-motor symptoms, supporting the potential utility in the clinical assessment of PD.
Journal IF-equivalent: 8.0 (OpenAlex 2-year mean citedness, value as of 2026-10-20, retrieved 2026-10-21; not the official Clarivate IF)Reference: Liu R, Kong Y, Fan L, Li L, yuan J, Zhang K, et al. Machine learning-based plasma proteomics discovers biomarkers for the early diagnosis of Parkinson’s disease. npj Parkinsons Dis. 2026 Nov 17 [Epub ahead of print]. doi:10.1038/s41531-026-01606-2.Checked: Abstract only