Digital Twin Brain: Generating Multitask Behavior from Connectomes for Personalized Therapy

Objective: This study introduces and validates a digital twin brain framework designed to translate an individual's brain connectome into predictions of multitask neurobehavioral dynamics and personalized functional modulations. Impact Statement: We introduce a novel 2-component architecture-where a hypernetwork personalizes a main network from an individual's connectome-establishing a mechanistic platform to simulate and design personalized interventions by directly linking connectomes to behavior. Introduction: Personalized psychiatry requires digital twin models that can predict functions across multiple domains, such as affective and cognitive processing, from an individual's unique neurobiology. However, existing models struggle to bridge the gap between brain structure and complex, multitask behavior, limiting their clinical utility. Methods: A hypernetwork uses an individual's resting-state connectome to generate parameters for a main recurrent neural network that simulates participant-specific behavioral and blood-oxygen-level-dependent (BOLD) time series across tasks. Leveraging the model's end-to-end architecture linking connectomes to behavior, we used gradient backpropagation to identify connectome manipulations designed to selectively modulate affective or cognitive functions. Results: Validated on 228 individuals, the model predicted behavioral choices with over 90% accuracy, reaction times (r > 0.85), and BOLD patterns (r = 0.84) with high fidelity. Crucially, in silico interventions successfully modulated targeted functions and reproduced realistic, interindividual variability in treatment effects arising from each person's baseline connectome. Conclusion: This digital twin brain system enables high-fidelity, in silico prediction and personalized modulation of complex neurobehavioral functions, advancing the potential for individualized psychiatric care.