Development of a Technology to Precisely Distinguish Schizophrenia and Bipolar Disorder through Psychiatric Genome Analysis
Background
The diagnostic framework of psychiatry was established in the late 19th century when Emil Kraepelin classified schizophrenia and bipolar disorder as distinct diseases. However, modern molecular genetics research has revealed that these two disorders share a significant portion of genetic vulnerability, suggesting that the two categories are not clearly separated but lie on the same genetic continuum. The overlapping nature of genetic variations acts as a barrier to accurate diagnosis in clinical settings. In particular, during the early stages of symptom onset, it is difficult to determine whether a patient belongs to either disorder based solely on clinical observation.
Conventional genome-wide analysis techniques have also shown clear limitations in resolving this issue. Traditional polygenic risk scores (PRS) use a one-to-one comparison method between a specific disease group and a healthy control group. For example, a predictive score derived from comparing schizophrenia patients with a control group does not provide useful information for differential diagnosis with bipolar disorder. When a patient's genetic variations increase the risk for both disorders, traditional predictive scores cannot yield a clear determination. This has led to a consistent demand for multidimensional predictive models that integrate genetic correlations across multiple disorders to distinguish specific conditions.
Key Findings
Dr. Wouter J. Peyrot from Vrije Universiteit Amsterdam and Professor Alkes L. Price from the Harvard T.H. Chan School of Public Health developed a statistical model that comprehensively evaluates the risk and genetic correlations of multiple disorders. Named the Differential Diagnosis–Polygenic Risk Score (DDx-PRS), this technique addresses the genetic covariance structure of various psychiatric disorders. The research team completed a Bayesian computational framework that combines existing predictive values calculated per disease to derive the probability of a patient belonging to schizophrenia, bipolar disorder, major depressive disorder, or the control group.
The research team conducted performance testing using genetic data from 11,460 patients registered with the Psychiatric Genomics Consortium (PGC). Analysis results showed that the area under the receiver operating characteristic curve (AUC), which represents the classification performance of a specific disease from other diseases and control groups, was 0.66 for schizophrenia (SCZ). Bipolar disorder (BIP) recorded an AUC of 0.64, and major depressive disorder (MDD) had an AUC of 0.59. The classification performance for distinguishing healthy controls was calculated at 0.68.
This statistical method has been evaluated as having stable calibration, indicating reliable predictive performance. This is due to the consistency between the risk probabilities derived by the model and the actual disease proportions in patient populations. Moreover, since the model completes its learning using only summarized statistical data without requiring additional clinical tuning data, it is easily applicable in clinical settings.
Significance and Prospects
The development of this predictive model has been praised for providing an objective genomic-based criterion for the subjective diagnosis of psychiatric disorders. It demonstrates the practical applicability of genetic information in the field of psychiatry, where physical diagnostic indicators have been limited. It is expected to become a supportive tool that helps clinicians make proactive treatment decisions, especially in the early stages of disease onset.
However, there are still areas for improvement before it can be used as a standalone diagnostic tool in clinical practice. To become a reliable standalone standard tool, the AUC value must exceed 0.8, but the results of this model do not reach that threshold. The research team plans to develop a composite algorithm that integrates genetic risk with environmental factors and behavioral symptom changes. Follow-up research should also include additional genetic data from non-European populations to ensure broader applicability across diverse population groups.
Nature Genetics, Published online: 20 August 2026; doi:10.1038/s41588-026-02684-xDifferential diagnosis–polygenic risk score (DDx-PRS) is a method to distinguish related psychiatric disorders from each other and from controls by modeling the variance–covariance structure of disorder liabilities and polygenic risk scores.
A major challenge for psychiatrists in clinical settings is distinguishing the diagnosis of patients who first exhibit psychiatric symptoms during adolescence. When patients visit the hospital with vague symptoms such as delusions or mood swings, it is not easy to immediately determine whether the condition is schizophrenia or bipolar disorder. Since the medications for these two disorders differ, diagnostic errors can lead to side effects or worsening symptoms.
Imagine a hypothetical scenario in which a patient's genome is extracted and analyzed using the diagnostic program. If the test results show a specific prediction of 72% probability for schizophrenia and 12% for bipolar disorder, medical staff can gain confidence in establishing a treatment direction. By preemptively prescribing schizophrenia medication, clinicians can prevent chronic progression and promote faster recovery. In this way, the genetic data model functions as an auxiliary indicator that enhances diagnostic accuracy in the clinic.