A novel potential biomarker panel to diagnose depression derived from big proteomic data

Background
There is still no clinical biomarker to diagnose depression. Given the complexity of a multifactorial disease like depression, a single biomarker is unlikely to capture the full heterogeneity of the disease and be applicable in clinical practice, mandating biomarker panels representing several biological targets.
Methods
We examined two proteomic datasets from the UK Biobank: the Cox dataset (N = 19,632) and the diagnostic dataset (N = 19,374). Cox proportional hazards regression modeling was used to identify potential biomarkers of depression within the Cox dataset, and subsequently the diagnostic accuracy of these candidate biomarkers was validated in the diagnostic dataset. Employing four distinct machine learning algorithms and LASSO regression model, we discovered the most effective biomarker panel for depression, assessing model performance through five-fold cross-validation and the area under receiver operating characteristic curve (AUC).
Results
Over a mean follow-up of 14 years, 46 plasma proteins were significantly associated with depression after adjusting for confounders. These depression-related proteins were involved in immune-related processes and pathways. When combined with traditional risk factors, the six blood protein biomarkers identified in this study achieved 75.4 % diagnostic accuracy for depression, which was similar to using 46 (maximum 74.9 %) and 2911 (maximum 75.9 %) proteins.
Conclusions
Our findings suggest the potential clinical use of proteomic biomarkers as complementary information for early and population-based detection of depression. With appropriate clinical and experimental validation, the identified depression-related proteins may be used as a biomarker panel for the screening and prediction of depression.