Dataset Information

Multi-site benchmark classification of major depressive disorder using machine learning on cortical and subcortical measures.

ABSTRACT: Machine learning (ML) techniques have gained popularity in the neuroimaging field due to their potential for classifying neuropsychiatric disorders. However, the diagnostic predictive power of the existing algorithms has been limited by small sample sizes, lack of representativeness, data leakage, and/or overfitting. Here, we overcome these limitations with the largest multi-site sample size to date (N = 5365) to provide a generalizable ML classification benchmark of major depressive disorder (MDD) using shallow linear and non-linear models. Leveraging brain measures from standardized ENIGMA analysis pipelines in FreeSurfer, we were able to classify MDD versus healthy controls (HC) with a balanced accuracy of around 62%. But after harmonizing the data, e.g., using ComBat, the balanced accuracy dropped to approximately 52%. Accuracy results close to random chance levels were also observed in stratified groups according to age of onset, antidepressant use, number of episodes and sex. Future studies incorporating higher dimensional brain imaging/phenotype features, and/or using more advanced machine and deep learning methods may yield more encouraging prospects.

SUBMITTER: Belov V

PROVIDER: S-EPMC10784593 | biostudies-literature | 2024 Jan

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Multi-site benchmark classification of major depressive disorder using machine learning on cortical and subcortical measures.

Belov Vladimir V Erwin-Grabner Tracy T Aghajani Moji M Aleman Andre A Amod Alyssa R AR Basgoze Zeynep Z Benedetti Francesco F Besteher Bianca B Bülow Robin R Ching Christopher R K CRK Connolly Colm G CG Cullen Kathryn K Davey Christopher G CG Dima Danai D Dols Annemiek A Evans Jennifer W JW Fu Cynthia H Y CHY Gonul Ali Saffet AS Gotlib Ian H IH Grabe Hans J HJ Groenewold Nynke N Hamilton J Paul JP Harrison Ben J BJ Ho Tiffany C TC Mwangi Benson B Jaworska Natalia N Jahanshad Neda N Klimes-Dougan Bonnie B Koopowitz Sheri-Michelle SM Lancaster Thomas T Li Meng M Linden David E J DEJ MacMaster Frank P FP Mehler David M A DMA Melloni Elisa E Mueller Bryon A BA Ojha Amar A Oudega Mardien L ML Penninx Brenda W J H BWJH Poletti Sara S Pomarol-Clotet Edith E Portella Maria J MJ Pozzi Elena E Reneman Liesbeth L Sacchet Matthew D MD Sämann Philipp G PG Schrantee Anouk A Sim Kang K Soares Jair C JC Stein Dan J DJ Thomopoulos Sophia I SI Uyar-Demir Aslihan A van der Wee Nic J A NJA van der Werff Steven J A SJA Völzke Henry H Whittle Sarah S Wittfeld Katharina K Wright Margaret J MJ Wu Mon-Ju MJ Yang Tony T TT Zarate Carlos C Veltman Dick J DJ Schmaal Lianne L Thompson Paul M PM Goya-Maldonado Roberto R

Scientific reports 20240111 1

Machine learning (ML) techniques have gained popularity in the neuroimaging field due to their potential for classifying neuropsychiatric disorders. However, the diagnostic predictive power of the existing algorithms has been limited by small sample sizes, lack of representativeness, data leakage, and/or overfitting. Here, we overcome these limitations with the largest multi-site sample size to date (N = 5365) to provide a generalizable ML classification benchmark of major depressive disorder (M ...[more]

PMID: 38212349

Similar Datasets

Project description:Background: Major Depressive Disorder (MDD) is a moderately heritable disorder with a high lifetime prevalence. At present, laboratory blood tests to support MDD diagnosis are not available. Methods: We used a classifier approach on blood gene expression profiles of a unique set of non-medicated subjects (MDD patients and controls) to select genes of which expression is predictive for disease status. To reveal blood gene expression changes related to MDD disease, we applied a powerful ex vivo stimulus to the blood, i.e. incubation with lipopolysaccharide (LPS; 10 ng/ml blood). Results: Based on LPS-stimulated blood gene expression using whole-genome microarrays in 42 subjects (primary cohort; 21 MDD patients (mean age 42.3 years), 21 healthy controls (mean age 41.9 years)), we identified a set of genes (CAPRIN1, CLEC4A, KRT23, MLC1, PLSCR1, PROK2, ZBTB16) that serves as a molecular signature of MDD. These findings were validated for the primary cohort using an independent quantitative PCR method (P = 0.007). The difference between depressive patients and controls was confirmed (P = 0.019) in a replication cohort of 13 patients with MDD (mean age 42.8 years) and 14 controls (mean age 45.6 years). The MDD-signature score comprised of expression levels of 7 genes could discriminate depressive patients from controls with sensitivity of 76.9% and specificity of 71.8%. Conclusions: We show for the first time that molecular analysis of stimulated blood cells can be used as an endophenotype for MDD diagnosis, which is a milestone in establishing biomarkers for neuropsychiatric disorders with moderate heritability in general. Our results may provide a new entry point for following and predicting treatment outcome, as well as prediction of severity and recurrence of MDD. In total, 33 MDD patients and 34 healthy controls were analyzed using basal gene expression in whole blood, and gene expression from whole blood that was stimulated with LPS for 5-6 h, using microarrays. Patients were arbitrarily selected from all patients to serve as primary cohort (nMDD = 21 (MDD01-MDD21); nControls = 21 (Con01-Con21)), or replication cohort (nMDD = 12 (MDD22-MDD35); nControls = 13 (Con22-Con37)) using microarrays. This submission does not include Samples CON21_LPS or CON30_LPS.

Dataset Information

Multi-site benchmark classification of major depressive disorder using machine learning on cortical and subcortical measures.

Publications

Multi-site benchmark classification of major depressive disorder using machine learning on cortical and subcortical measures.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets