Unveiling schizophrenia’s neural and mental signatures with machine learning

Neurocognitive profiles distinguish between SCZ and HCS. Credit: Chen et al. (Nature Mental Health, 2026).

Schizophrenia is a severe and often highly debilitating psychiatric disorder characterized by distorted emotions, thinking patterns and altered perceptions of reality, as well as mental impairments. This disorder typically emerges in late adolescence or early adulthood and requires lifelong treatment with antipsychotic medications.

The symptoms of schizophrenia include hallucinations and false or irrational beliefs about oneself and the world (i.e., delusions), as well as memory impairments, attention deficits and learning difficulties. Currently, schizophrenia is diagnosed by psychiatrists or other trained mental health specialists via comprehensive interviews and clinical assessments.

Identifying neural and mental signatures of schizophrenia could help to develop additional tools to diagnose and monitor the disorder or better plan the treatment of individual patients. In a recent paper published in Nature Mental Health, researchers at University of Washington, University of California San Diego, the VA San Diego Healthcare System and other institutes showed that machine learning models can accurately distinguish between people diagnosed with schizophrenia and others with no mental health disorders, by analyzing data collected during neurocognitive tests.

“The development of neurocognitive biomarkers for schizophrenia has relied on lengthy test batteries that are infeasible to deploy in clinical settings,” wrote Robert Y. Chen, Tiffany A. Greenwood and their colleagues in their paper. “Using machine learning, we sought to identify a subset of neurocognitive domains that could distinguish between patients with schizophrenia and healthy comparison subjects (HCS).”

Analyzing data collected from patients with schizophrenia

The researchers developed a machine learning model, a computational model that can rapidly analyze large amounts of data and learn to recognize patterns in this data or predict specific outcomes. This model was trained on neurocognitive data that were collected from 559 people diagnosed with schizophrenia or schizoaffective disorder (which combines some symptoms of schizophrenia with a mood disorder, like depression or bipolar disorder) and 745 people with no known mental health disorders.

The individuals from which the data was collected had completed 15 established tests that assess different thinking and brain functions. The machine learning model developed by the researchers learned patterns in this data that could be used to distinguish patients diagnosed with schizophrenia or schizoaffective disorder from people with no mental health disorders.

“We developed a machine learning model that could accurately separate patients with schizophrenia from HCS and was replicated in an independent cohort,” wrote Chen, Greenwood and their colleagues. “Recursive feature elimination revealed that just two neurocognitive domains—verbal learning and emotion identification—were sufficient to achieve the same classification accuracy.”

Towards more reliable diagnostic tools for schizophrenia

Chen, Greenwood and their colleagues showed that their model could distinguish between people diagnosed with schizophrenia and those who received no neuropsychiatric diagnoses with remarkable accuracy. Moreover, they showed that two specific mental capabilities, namely verbal learning (i.e., the ability to learn and memorize words) and emotion identification (i.e., recognizing and interpreting other’s emotions) were particularly crucial for the prediction of schizophrenia.

When the model focused only on these two cognitive abilities, it achieved an accuracy comparable to the one attained when exploring all the mental capabilities assessed during the neurocognitive tests. This suggests that deficits in these two areas are most predictive of schizophrenia and schizoaffective disorder.

“These findings support a ‘less-is-more’ approach to efficient neurocognitive profiling across the schizophreniform spectrum and highlight what may be the most impaired neurocognitive domains in this debilitating disorder,” wrote the authors.

In the future, the results of this recent study could inform the development of new tools that could simplify the diagnosis of schizophrenia and schizoaffective disorder or help mental health practitioners to devise treatment plans that best support individual patients. In addition, they could inform further research aimed at uncovering the most predictive cognitive and neural signatures of specific psychiatric disorders using machine learning.

© 2026 Science X Network

An individual’s sex may affect the course of bipolar disorder and schizophrenia