Precise stratification of patients into homogeneous disease subgroups could address the heterogeneity of phenotypes and enhance understanding of the pathophysiology underlying specific subtypes. Existing literature on subtyping patients with major depressive disorder (MDD) mainly utilized clinical features only. Genomic and imaging data may improve subtyping, but advanced methods are required due to the high dimensionality of features.

We propose a novel disease subtyping framework for MDD by integrating brain structural features, genotype-predicted expression levels in brain tissues, and clinical features. Using a multi-view biclustering approach, we classify patients into clinically and biologically homogeneous subgroups. Additionally, we propose approaches to identify causally relevant genes for clustering.

We verified the reliability of the subtyping model by internal and external validation. High prediction strengths (PS) (average PS: 0.896, minimum: 0.854), a measure of generalizability of the derived clusters in independent datasets, support the validity of our approach. External validation using patient outcome variables (treatment response and hospitalization risks) confirmed the clinical relevance of the identified subgroups. Furthermore, subtype-defining genes overlapped with known susceptibility genes for MDD and were involved in relevant biological pathways. In addition, drug repositioning analysis based on these genes prioritized promising candidates for subtype-specific treatments.

Our approach successfully stratified MDD patients into subgroups with distinct clinical prognoses. The identification of biologically and clinically meaningful subtypes may enable more personalized treatment strategies. This study also provides a framework for disease subtyping that can be extended to other complex disorders.