2026 ECNP Congress

PS01-1083

Differential diagnosis of melancholic depression and bipolar disorder using machine learning–assisted proximity extension assay-based plasma proteomic analysis

Reyhan Nur Babalıoğlu — B. Karacicek, B. Ozturk, B.I. Arioz, B. Cavusoglu, B. Verim, S. Balac, S. Gurkas, Z.G. Daglar, D. Ceylan, E. Bora, C.E. Teunissen, S. Genc, P. Keskinoglu · Koç University, Neuroscience, Istanbul, Türkiye

2026 ECNP Congress

Background

Major depressive disorder (MDD) and bipolar disorder (BD) are both common and clinically challenging mood disorders [1,2]. In everyday practice, distinguishing between them is not always straightforward, especially since depressive episodes are often the first presentation of BD. Because of this, many patients initially receive an MDD diagnosis, and the correct diagnosis may be delayed for several years [3]. This delay can have important treatment implications. Although neuroimaging studies have provided some insights into differences between these disorders, they are not easily translated into routine clinical use. Peripheral biomarkers, particularly those derived from proteomic analyses, have been proposed as a more accessible alternative [1,2]. Still, most reported markers seem to reflect shared biological pathways rather than disorder-specific features [1,2]. One possible way forward is to combine high-dimensional proteomic data with machine learning methods to identify more specific patterns [4,5]. In this study, we explored whether plasma proteomic profiles could help distinguish melancholic BD from MDD.

Methods

This case–control study included 67 participants (23 BD, 20 MDD, and 24 healthy controls). Diagnoses were made according to DSM-IV-TR criteria and confirmed with SCID-5 interviews. All patients were euthymic at the time of assessment, which allowed us to minimize the effects of acute symptoms. Melancholic features during previous depressive episodes were assessed using the Sydney Melancholia Prototype Index. Clinical measures included HDRS, YMRS, and HCL-32. Plasma proteins were measured using the Olink Target 96 Neurology panel based on proximity extension assay (PEA) technology, and results were expressed as log2-normalized NPX values. Group differences were examined using Kruskal–Wallis tests with a conservative threshold (p ≤ 0.001). For classification, we tested several machine learning approaches, including Random Forest, XGBoost, Logistic Regression, Support Vector Machines, and Naive Bayes. The dataset was split into training and test sets (70/30), and performance was evaluated using 5×5 nested cross-validation and area under the ROC curve (AUC).

Results

The groups differed in age, body mass index, years of education, and HCL-32 scores. However, HDRS and YMRS scores were comparable, suggesting that symptom severity was similar at the time of sampling. When we visualized the data using t-SNE, the BD group appeared to separate quite clearly from both MDD and controls. In univariate analyses, 21 proteins differed between BD and MDD. Among these, eight proteins (BMP4, CD38, CTSC, HAGH, LXN, MATN3, NCAN, and PLXNB1) consistently distinguished BD from both comparison groups. In the machine learning analyses, models using all proteins reached an AUC of up to 0.868. After reducing the feature set, a five-protein panel(LXN, HAGH, MATN3, PLXNB1, CTSC) performed better, with an AUC of 0.905.

Conclusions

Taken together, these findings suggest that combining plasma proteomics with machine learning may help improve the differentiation of BD and MDD. The results point toward a more distinct peripheral biological profile in BD. Interestingly, since symptom scores were similar across groups, these differences may reflect more stable, trait-like characteristics rather than current clinical state. Although the sample size is relatively modest, the identified protein panel seems promising and deserves further investigation in larger and independent samples.

References

  1. Gopalakrishnan, R., Wang, Y., Kapczinski, F., Frey, B. N., & Wollenhaupt-Aguiar, B. (2025). Peripheral protein inflammatory biomarkers in bipolar disorder and major depressive disorder: A systematic review and meta-analysis. Journal of affective disorders, 376, 149–168. https://doi.org/10.1016/j.jad.2025.01.150
  2. Poletti, S., Mazza, M. G., & Benedetti, F. (2024). Inflammatory mediators in major depression and bipolar disorder. Translational psychiatry, 14(1), 247. https://doi.org/10.1038/s41398-024-02921-z
  3. Martinuzzi, E., Barbosa, S., Courtet, P., Olié, E., Guillaume, S., Ibrahim, E. C., Daoudlarian, D., Davidovic, L., Glaichenhaus, N., & Belzeaux, R. (2021). Blood cytokines differentiate bipolar disorder and major depressive disorder during a major depressive episode: Initial discovery and independent sample replication. Brain, behavior, & immunity - health, 13, 100232. https://doi.org/10.1016/j.bbih.2021.100232
  4. Tomasik, J., Han, S. Y. S., Barton-Owen, G., Mirea, D. M., Martin-Key, N. A., Rustogi, N., Lago, S. G., Olmert, T., Cooper, J. D., Ozcan, S., Eljasz, P., Thomas, G., Tuytten, R., Metcalfe, T., Schei, T. S., Farrag, L. P., Friend, L. V., Bell, E., Cowell, D., & Bahn, S. (2021). A machine learning algorithm to differentiate bipolar disorder from major depressive disorder using an online mental health questionnaire and blood biomarker data. Translational psychiatry, 11(1), 41. https://doi.org/10.1038/s41398-020-01181-x
  5. Poletti, S., Vai, B., Mazza, M. G., Zanardi, R., Lorenzi, C., Calesella, F., Cazzetta, S., Branchi, I., Colombo, C., Furlan, R., & Benedetti, F. (2021). A peripheral inflammatory signature discriminates bipolar from unipolar depression: A machine learning approach. Progress in neuro-psychopharmacology & biological psychiatry, 105, 110136. https://doi.org/10.1016/j.pnpbp.2020.110136