A Multi-Cohort Stable Gene Panel for Parkinson's Disease Detection from Blood Genomics Parkinson Hastali?ginin Kan Genomi?ginden Tespiti için Çok Kohortlu Kararli Gen Paneli
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636775
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: blood biomarker, cross-cohort validation, gene expression, machine learning, Parkinson's disease
- Gazi Üniversitesi Adresli: Evet
Özet
This study proposes Heterogeneity-Penalized Cross-Cohort Stability Scoring (H-CSP) to distinguish Parkinson's disease (PD) patients from healthy controls using blood. Four GEO microarray datasets (GSE6613, GSE57475, GSE72267, GSE99039; 754 samples, 4 platforms) were used. H-CSP penalizes platform-inconsistent gene expressions via the I2 statistic. Under a Leave-One-Dataset-Out (LODO) protocol with cohort-independent normalization, a 37-gene consensus panel reached a cross-platform discrimination of AUC= 0.810±0.074 (sensitivity 0.76, specificity 0.72) with L2-regularized logistic regression. Under a fully nested, leakage-free evaluation H-CSP achieved AUC = 0.626 ± 0.081, within the performance range of recent single-cohort panels (AUC≈ 0.60) assessed under the same protocol. Its distinctive contribution is reducing cross-cohort performance variance, yielding a consistent gene selection. CACNA1D, RGS2, PTGDS and BCL2L2 are linked to PD pathogenesis.