Diagnosis of major depressive disorder by combining multimodal information from heart rate dynamics and serum proteomics using machine-learning algorithm

• Published in: Progress in neuro-psychopharmacology & biological psychiatry Vol. 76; pp. 65 - 71
• Main Authors: Kim, Eun Young, Lee, Min Young, Kim, Se Hyun, Ha, Kyooseob, Kim, Kwang Pyo, Ahn, Yong Min
• Format: Journal Article
• Language: English
• Published: England Elsevier Inc 02.06.2017
• Subjects: Adult; Biomarker; Biomarkers; Depressive Disorder, Major - blood; Depressive Disorder, Major - diagnosis; Depressive Disorder, Major - physiopathology; Female; Heart Rate - physiology; Heart rate variability; Humans; Machine Learning; Major depressive disorder; Middle Aged; Proteome - metabolism; Proteomics; Biomarker; Heart rate variability; Machine-learning; Major depressive disorder; Proteomics
• DOI: 10.1016/j.pnpbp.2017.02.014
• ISSN: 0278-5846

Major depressive disorder (MDD) is a systemic and multifactorial disorder that involves abnormalities in multiple biochemical pathways and the autonomic nervous system. This study applied a machine-learning method to classify MDD and control groups by incorporating data from serum proteomic analysis and heart rate variability (HRV) analysis for the identification of novel peripheral biomarkers. The study subjects consisted of 25 drug-free female MDD patients and 25 age- and sex-matched healthy controls. First, quantitative serum proteome profiles were analyzed by liquid chromatography-tandem mass spectrometry using pooled serum samples from 10 patients and 10 controls. Next, candidate proteins were quantified with multiple reaction monitoring (MRM) in 50 subjects. We also analyzed 22 linear and nonlinear HRV parameters in 50 subjects. Finally, we identified a combined biomarker panel consisting of proteins and HRV indexes using a support vector machine with recursive feature elimination. A separation between MDD and control groups was achieved using five parameters (apolipoprotein B, group-specific component, ceruloplasmin, RMSSD, and SampEn) at 80.1% classification accuracy. A combination of HRV and proteomic data achieved better classification accuracy. A high classification accuracy can be achieved by combining multimodal information from heart rate dynamics and serum proteomics in MDD. Our approach can be helpful for accurate clinical diagnosis of MDD. Further studies using larger, independent cohorts are needed to verify the role of these candidate biomarkers for MDD diagnosis. •Biomarkers for depression were identified using machine learning.•A combined biomarker panel consisting of proteins and HRV indexes was developed.•Classification accuracy of 80.1% was achieved by combining HRV and proteomic data.•Biomarkers: apolipoprotein B, group-specific component, ceruloplasmin, RMSSD, SampEn