A Dual Prediction Error Populations Model of Schizophrenia Spectrum Disorders: Bridging NMDA Hypofunction and Gain Hyperfunction

BACKGROUND AND HYPOTHESIS: Schizophrenia spectrum disorders (SSDs) span heterogeneous clinical states, from rapidly fluctuating psychosis to rigid, systematized delusions.1 NMDA receptor hypofunction on inhibitory interneurons,2-4 dopaminergic dysregulation,5 and abnormalities of volatility estimation6-10 or precision weighting11-13 have all been implicated, yet a framework linking these mechanisms to spectrum-level heterogeneity remains incomplete. STUDY DESIGN: We developed a biologically plausible predictive-coding model with separable positive and negative prediction error (PE) populations. For clarity, the main text presents a one-dimensional linear Kalman filter. STUDY RESULTS: NMDA hypofunction selectively weakened negative PE, thereby reducing disconfirmatory updating and biasing inference toward prior-consistent interpretations. Gain hyperfunction amplified precision-weighted updating and produced observation-dominated instability. Their interaction reproduced attenuated mismatch responses, reduced illusion susceptibility, and a bias-volatility plane spanning rigid and unstable delusional dynamics. CONCLUSIONS: The model suggests that SSD heterogeneity can be understood as an interaction between sign-selective bias and abnormal precision weighting. This offers a circuit-to-symptom framework and motivates stratification by the asymmetry and overall gain of PE responses.