Multimodal Mechanisms of Mood Stabilization: A Computational Model of Antiepileptics Action in Bipolar Disorder Progression

Background: Bipolar disorder features mood episodes that tend to become more frequent and severe, a pattern often explained by sensitization or kindling. Although lithium and valproate reliably prevent episodes, their precise actions—ranging from neuroprotection to restoring the balance between excitation and inhibition—are not fully reconciled. Computational modeling makes it possible to test these mechanisms over extended time frames. Methods: Using a recurrent neural-network pruning model that can display both depressive and bipolar dynamics, we created a "bipolar-classic" network (80 % synaptic sparsity with an inhibitory bias). The network underwent fifteen alternately triggered manic- and depressive-like episodes. Each episode produced permanent scarring, excitotoxic loss, cross-pole interactions, and progressive lowering of activation thresholds. Eight treatment scenarios were simulated: four single mechanisms—neuroprotection, inhibitory rebalancing, instability damping, and anti-sensitization—and four combinations. Ten random seeds were run per condition. Outcomes were compared with clinical prophylaxis patterns through a weighted composite score and Bayesian model comparison. Results: Instability damping produced the largest reduction in total episodes (mean 8.6 ± 0.5) versus untreated controls (14.6 ± 0.7), mainly by nearly abolishing manic events. However, dual mechanisms, especially neuroprotection plus inhibitory rebalancing, yielded the best restoration of excitation–inhibition balance (final ratio about 0.52 versus the initial 0.11) while still generating realistic episode sequences. Combination treatments also showed the greatest structural recovery, with very large effect sizes (Cohen's d up to –11.75 for the E/I ratio) relative to control. Conclusions: The simulations support a multimodal explanation for mood-stabilizer efficacy. Synergy between neuroprotection and inhibitory rebalancing appears most consistent with clinical observations, whereas pure damping may represent a secondary consequence rather than a primary drug action. These findings argue for early, multifaceted intervention to counteract kindling-like progression in bipolar disorder and demonstrate how computational models can guide the search for convergent therapeutic pathways.