Individual-First Brain State Modeling Reconciles Personalized Prediction with Generalizable Mechanistic Interpretation in Precision Psychiatry

Neuropsychiatric heterogeneity creates a fundamental trade-off between personalized prediction and generalizable mechanistic interpretation. Conventional brain metastate models enforce shared state repertoires, inherently suppressing clinically meaningful individual variation, whereas individualized brain states lack direct correspondence across participants, limiting cross-subject comparison. We present Individual-specific Brain State Space (IBSS), an interpretable individual-first framework that estimates brain states separately for each participant and models their spatiotemporal organization for individualized prediction. We then cluster similar individualized states across participants into shared metastates, thereby reconstructing a unified metastate space for generalizable neurodynamic interpretation. Across four large-scale fMRI cohorts (HCP-task, ABIDE-I/II, and DIRECT-II), the IBSS-based graph attention network achieved state-of-the-art performance in cognitive decoding and ASD/MDD identification, yielding accuracy gains of 16.66-36.69% over conventional group-first metastate approaches. These gains were primarily attributable to individualized state configurations rather than model complexity. Metastate reconstruction further revealed reproducible, symptom-linked neurodynamic patterns, including frontoparietal control network hyper-engagement in ASD and distinct default mode-limbic dynamics related to depressive and anxiety symptoms in MDD. This paradigm bridges personalized diagnostic prediction and shared neurodynamic interpretation, enabling robust heterogeneity characterization for precision psychiatry.