Bipolar disorder (BD) is a mood disorder characterized by the occurrence of depressive episodes alternating with episodes of elevated mood (known as mania). There is also an increased risk of other medical comorbidities.

This work uses a systems biology approach to compare BD treated patients with healthy controls (HCs), integrating proteomics and metabolomics data using partial correlation analysis in order to observe the interactions between altered proteins and metabolites, as well as proposing a potential metabolic signature panel for the disease.

Data integration between proteomics and metabolomics was performed using GC-MS data and label-free proteomics from the same individuals (N = 13; 5 BD, 8 HC) using generalized canonical correlation analysis and partial correlation analysis, and then building a correlation network between metabolites and proteins. Ridge-logistic regression models were developed to stratify between BD and HC groups using an extended metabolomics dataset (N = 28; 14 BD, 14 HC), applying a recursive feature elimination for the optimal selection of the metabolites.

Network analysis demonstrated links between proteins and metabolites, pointing to possible alterations in hemostasis of BD patients. Ridge-logistic regression model indicated a molecular signature comprising 9 metabolites, with an area under the receiver operating characteristic curve (AUROC) of 0.833 (95% CI 0.817-0.914).

From our results, we conclude that several metabolic processes are related to BD, which can be considered as a multi-system disorder. We also demonstrate the feasibility of partial correlation analysis for integration of proteomics and metabolomics data in a case-control study setting.