Summary: For the first time, lab-grown brain organoids have revealed how neurons misfire in schizophrenia and bipolar disorder. By applying machine learning to the organoids’ electrical activity, researchers detected distinct electrophysiological signatures that distinguish patient-derived tissue from healthy controls with over 90% accuracy after stimulation.
Gentle electrical stimulation amplified subtle differences in neural activity, exposing condition-specific neuro-signatures. These findings point to new diagnostic approaches and preclinical platforms for testing psychiatric treatments, with the potential to reduce trial-and-error prescribing.
Key Facts
- Distinct Biomarkers: Organoids displayed unique neural firing patterns linked to schizophrenia and bipolar disorder.
- High Accuracy: Machine learning classified organoids with up to 92% accuracy after electrical stimulation; other models reached 95.8% in two-dimensional cultures.
- Clinical Potential: The method could support more objective diagnostics and personalized drug testing for psychiatric conditions.
Source: Johns Hopkins University
Pea-sized brain organoids grown in the laboratory have, for the first time, exposed characteristic ways neurons may malfunction in schizophrenia and bipolar disorder—two complex psychiatric illnesses that lack objective molecular biomarkers and rely on clinical assessment for diagnosis.
These results could help clinicians reduce lengthy trial-and-error treatment approaches by offering new laboratory tests to guide diagnosis and drug selection for individuals with mental illness.
The study appears in APL Bioengineering.
“Schizophrenia and bipolar disorder are very hard to diagnose because no specific brain region or single molecular marker reliably differentiates patients from healthy people,” said Annie Kathuria, a biomedical engineer at Johns Hopkins University who led the research. “Our approach aims to read the electrical language of developing brain tissue to reveal condition-specific signatures.”
Kathuria and colleagues reprogrammed blood and skin cells from individuals with schizophrenia, bipolar disorder, and healthy controls into induced pluripotent stem cells, then guided those cells to form cerebral organoids—miniature, simplified versions of the human brain packed with multiple neural cell types.
To monitor network activity, the team placed organoids on microchips fitted with multi-electrode arrays (MEAs), which record electrical signals much like a scaled-down electroencephalogram (EEG). They tracked spontaneous neural firing during development and after applying mild electrical stimulation designed to evoke additional activity.
Using a machine learning pipeline optimized for high-dimensional electrophysiology data, the researchers extracted features reflecting channel-specific network activity, spike timing, and other metrics. Those features served as biomarkers that could separate organoids derived from patients with schizophrenia and bipolar disorder from controls.
At baseline, organoid-derived signatures distinguished patient and control samples with approximately 83% accuracy. After brief, subtle electrical stimulation, classification accuracy rose to about 92% in the organoids. In complementary two-dimensional interneuron cultures, a Support Vector Machine classifier achieved 95.8% accuracy in distinguishing schizophrenia from controls under both baseline and stimulated conditions.
The stimulation-sensitive patterns included complex combinations of firing spikes and timing changes across multiple channels and intervals, creating distinct electrophysiological fingerprints for each disorder. These fingerprints made it possible to classify samples reliably and to pinpoint which aspects of network behavior differed between cohorts.
Fully developed organoids in the study reached roughly three millimeters in diameter and contained diverse neural cell types typical of the prefrontal cortex, a brain region involved in higher cognitive functions. The tissue also displayed myelination—cellular insulation that supports efficient electrical signaling—further supporting the organoids’ relevance as functional models.
The study used samples from 12 individuals, a modest cohort, but the authors emphasize the translational potential of their approach. By combining patient-derived tissue, MEA recording, and machine learning, the platform could evolve into a testbed for screening psychiatric drugs and optimizing dosages tailored to an individual’s neural phenotype.
Kathuria’s team is collaborating with neurosurgeons, psychiatrists, and neuroscientists at the Johns Hopkins School of Medicine to expand patient sampling and to evaluate how drug responses in organoids correspond to clinical effectiveness. The long-term goal is to reduce the months-long process of finding effective medications and to lower the proportion of patients who remain treatment-resistant.
“Many prescriptions today rely on trial and error. If organoids can predict which drugs restore healthy electrical patterns, clinicians may be able to select effective therapies more quickly,” Kathuria said. “For example, a significant share of patients are resistant to common antipsychotics; an organoid-based screen could help identify alternative treatments sooner.”
About this mental health and neuroscience research news
Author: Hannah Robbins
Source: Johns Hopkins University
Contact: Hannah Robbins – Johns Hopkins University
Image: The image is credited to Neuroscience News
Original Research: Open access. “Machine learning-enabled detection of electrophysiological signatures in iPSC-derived models of schizophrenia and bipolar disorder” by Annie Kathuria et al., APL Bioengineering. DOI reference available in the published article.
Abstract
Machine learning-enabled detection of electrophysiological signatures in iPSC-derived models of schizophrenia and bipolar disorder
Neuropsychiatric disorders such as schizophrenia (SCZ) and bipolar disorder (BPD) remain difficult to diagnose because objective biomarkers are lacking; current assessments depend largely on subjective clinical evaluations.
This study presents a computational pipeline to identify disease-specific electrophysiological signatures from multi-electrode array recordings of patient-derived cerebral organoids and two-dimensional interneuron cultures.
A Support Vector Machine classifier optimized for high-dimensional data achieved 95.8% accuracy in distinguishing SCZ from control samples in two-dimensional cultures under both baseline and post-electrical-stimulation conditions using extracted electrophysiological features. In cerebral organoids, classification accuracy improved from 83.3% at baseline to 91.6% after stimulation, enabling robust separation of control, SCZ, and BPD cohorts.
Key discriminative features included channel-specific measures of network activity, and post-stimulation recordings significantly enhanced classification performance, particularly for bipolar disorder. These results highlight the potential of MEA-based functional phenotyping combined with machine learning to reveal stimulation-sensitive electrophysiological biomarkers, paving the way toward more objective diagnosis and personalized treatment strategies for neuropsychiatric disorders.