Excessive Synaptic Pruning Induces Network Fragility That Is Rescued by Simulated Neuroplasticity: A Computational Model of Vulnerability and Recovery in Major Depressive Disorder

Background: Major depressive disorder (MDD) is highly heritable and polygenic. Recent work points to abnormal developmental synaptic pruning—rather than primary glutamatergic failure—as the central liability, with later-life plasticity limits and reduced cognitive reserve amplifying risk under stress. Although computational studies have examined pruning in other psychiatric settings, few models capture depression-specific fragility or test whether a treatment that boosts synaptogenesis can reverse that state. Methods: We built an over-parameterised feed-forward network (~400 000 weights) and trained it on a noisy four-class task, representing exuberant early connectivity. Excessive adolescent pruning was mimicked by removing 95 % of weights in every layer. Robustness was measured on clean and perturbed inputs. To model rapid-acting antidepressants, half of the deleted connections were reinstated at small random strengths, after which the network was fine-tuned. Results: The dense model reached near-perfect accuracy with strong noise tolerance. After pruning, clean-set accuracy fell to 50.8 % and dropped further under noise (43.1 %), revealing marked fragility. Regrowth plus brief retraining restored baseline accuracy on clean data and recovered 84–98 % accuracy under perturbations, even though the final network remained ~47 % sparse. Conclusions: Excessive connectivity loss created a threshold-like collapse in performance that parallels stress-induced relapse in MDD. Limited synaptic regrowth—a stand-in for ketamine-induced plasticity—largely reversed this collapse without returning the network to its original density. The simulation supports the pruning-mediated plasticity-deficit hypothesis and suggests that therapies which reopen structural plasticity windows can compensate for irreversible developmental losses.