Identification of transdiagnostic psychiatric disorder subtypes using unsupervised learning

Area of Science:

• Psychiatry
• Genetics
• Data Science

Background:

• Psychiatric disorders exhibit significant heterogeneity in symptoms and progression.
• Current diagnostic systems do not fully capture underlying molecular etiologies.
• This heterogeneity complicates accurate diagnosis and targeted treatment development.

Purpose of the Study:

• To identify patient clusters with shared clinical and genetic characteristics.
• To explore potential therapeutic benefits of grouping patients transdiagnostically.
• To better understand the molecular and symptomatic diversity within psychiatric conditions.

Main Methods:

• Applied high-dimensional data clustering to deep clinical data from a large sample.
• Analyzed genetic data, including polygenic scores (PGS) and family history, across identified clusters.
• Validated cluster findings in an independent patient sample.

Main Results:

• Identified five diagnostically mixed clusters, ordered by severity, with distinct clinical profiles.
• Cluster 4, comprising mostly psychotic disorder patients, showed highest severity and medication load.
• Genetic analysis revealed significant differences between the least and most severe clusters (AUC=81%), highlighting shared genetic risk factors.

Conclusions:

• Psychiatric disorders comprise heterogeneous subtypes with shared molecular and symptomatic features.
• Transdiagnostic clustering advances understanding of psychiatric heterogeneity.
• Findings support the development of personalized treatment strategies based on identified patient subtypes.