<p>Bipolar disorder (BD) is a chronic mental disorder, but its neural mechanism and treatment through non-invasive brain stimulation (NIBS) remain unclear. The whole-brain dynamic modeling integrated with neural activity data allows for systematic simulation of neuromodulation and can provide potential strategy to restore the abnormal brain networks. Here, we first applied a multilayer network method to study the dynamic integration of brain functional networks in BD patients at resting state. Then, we used a Hopf bifurcation model to construct the whole-brain network dynamic model and evaluated the effectiveness of neuromodulation in restoring the abnormal brain networks of BD patients. We found that the dynamic integration is significantly lower in BD patients (<i>n</i> = 49) relative to healthy controls (HC, <i>n</i> = 50) and shows an inverted U-shaped quadratic relationship with the mania severity. Using the whole-brain dynamic model, we observed that neuromodulation targeting specific brain regions can effectively restore the abnormal dynamic integration of BD patients to the normal level of HCs. Crucially, BD patients with mild and extremely severe mania have different degrees of dynamic integration reduction, and they also have heterogeneous optimal regions targeted by neuromodulation. Our results provide guidance for the qualitative analysis of the neuromodulation on abnormal brain networks and the development of personalized NIBS treatment strategies.</p>