Simulation-based neural dynamics of stress-related negative cognitive bias in depressive symptoms: a dynamical EEG/ERP framework

Introduction This simulation-based computational study develops a non-linear state-space framework for stress-related negative cognitive bias and multivariate EEG/ERP observations. Methods A three-state model represents negative bias, cognitive control, and arousal, while symptom load and normalized model stress input reshape a quartic bias potential. N2, P3, late positive potential, frontal theta, frontal alpha asymmetry (FAA), and frontoparietal phase locking provide complementary observation channels. Additional checks distinguish the non-linear benchmark from additive and interaction-only competing generators, examine equilibria of the full coupled system, and perturb the observation model. Results We establish bounded stochastic dynamics, a conditional scalar multistability boundary, stable-branch stress sensitivity, and a sufficient local state-observability condition. A seeded benchmark with 180 simulated participants yields model-implied stress-by-symptom slopes of 66.83 ms for response time, 1.12 μV for LPP, 0.775 dB for frontal theta, 0.133 for FAA, and 0.0769 for frontoparietal PLV. Under a non-linear generating process, mean held-out RMSE was 0.060 for the pre-specified quartic candidate vs. 0.436 for the interaction-only candidate. State-reconstruction RMSE increased from 0.152 under the nominal loading matrix to 0.232 under 15% loading perturbations. Discussion These results demonstrate implementation consistency and characterize model sensitivity; they are not clinical effect estimates or evidence of empirical parameter identifiability.