Integrating resting-state functional connectivity, mismatch negativity, and auditory steady-state responses reveals convergent neurophysiological biomarkers of schizophrenia

Cogn Neurodyn. 2026 Dec;20(1):195. doi: 10.1007/s11571-026-10561-w. Epub 2026 Oct 6.

ABSTRACT

Schizophrenia is a heterogeneous neuropsychiatric disorder characterized by abnormalities in large-scale brain connectivity, auditory prediction-error processing, and cortical gamma-band synchronization. Although resting-state electroencephalography (EEG), mismatch negativity (MMN), and auditory steady-state responses (ASSRs) have each been proposed as candidate biomarkers, they are often investigated separately, limiting the characterization of convergent neurophysiological abnormalities. This study evaluated whether integrating these complementary EEG paradigms could improve the biological characterization and predictive classification of schizophrenia within a unified multi-paradigm electrophysiological framework. A retrospective secondary analysis was performed using the publicly available African Schizophrenia EEG Dataset (ASZED), comprising 153 participants (76 individuals with schizophrenia and 77 healthy controls). Resting-state functional connectivity was quantified using the weighted Phase Lag Index and graph-theoretical analysis, while MMN peak amplitude, mean amplitude, latency, and 40-Hz ASSR gamma-band evoked power and inter-trial phase coherence were extracted as stimulus-evoked biomarkers. Multimodal classification was performed on 145 participants with complete resting-state, MMN, and ASSR recordings using a Graph Attention Network-based framework. Model interpretability was evaluated using SHAP and GNNExplainer, and predictive performance was assessed using nested stratified five-fold cross-validation with permutation testing. Participants with schizophrenia exhibited disrupted resting-state functional network organization, attenuated MMN responses, and impaired gamma-band synchronization compared with healthy controls. The integrated framework achieved the highest internally validated classification performance, with a balanced accuracy of 91.8% and an area under the receiver operating characteristic curve (AUC) of 0.96, exceeding the performance of the individual resting-state, MMN, and ASSR models within the matched complete-case cohort. Explainability analyses identified global efficiency, MMN peak amplitude, and ASSR inter-trial phase coherence as the most influential predictive features across validation folds. These findings indicate that resting-state connectivity, MMN, and ASSR provide complementary electrophysiological information that can be integrated into a coherent predictive representation of schizophrenia-related brain dysfunction. The findings should be interpreted as internally validated associations within a single public dataset rather than as evidence of clinically validated diagnostic biomarkers. Independent external validation and longitudinal studies will be required to determine whether these multi-paradigm EEG signatures have broader prognostic, treatment-response, or translational utility.

PMID:42846021 | PMC:PMC13642884 | DOI:10.1007/s11571-026-10561-w