Secretome-Driven Machine Learning for Prioritization of Blood-Based Biomarker Panels in Major Depressive Disorder

Major Depressive Disorder (MDD) remains a pressing global health challenge due to the absence of objective and scalable diagnostic tools. Because direct molecular sampling of the living brain is impossible, research efforts increasingly rely on peripheral blood as a window into disease biology. In this study, whole-blood transcriptomic datasets were systematically analyzed to identify biomarkers with direct translational potential. By focusing on genes encoding secretory proteins detectable in peripheral blood, the study prioritizes candidates that may be more amenable to future translational evaluation. Differential expression analysis revealed 20 such genes, all of which have commercially available ELISA kits, indicating assay feasibility and facilitating future protein-level validation in clinical studies. A rigorous, leakage-free machine learning framework incorporating nested cross-validation, feature sparsity, and probabilistic calibration was implemented to assess diagnostic performance. Four models, Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), and XGBoost (XGB) were compared, with LR emerging as the most robust classifier due to superior stability, interpretability, and reproducibility. LR achieved strong discriminative performance in both internal validation (AUC = 0.759, AUPRC = 0.867) and external datasets (AUC = 0.664, AUPRC = 0.704). Network and correlation analyses further confirmed that the identified genes are central to immune-inflammatory pathways known to play critical roles in MDD pathophysiology. Notably, novel candidates including PF4V1, FOLR3, and GZMK, along with 10 newly implicated and 6 previously reported genes, were computationally validated. These findings present a computationally prioritized and internally validated panel of candidate secretory blood biomarkers for MDD, supporting future development of minimally invasive diagnostic assays pending further assessment.