Traditional toxicity detection focuses on individuals and often overlooks coordinated group behaviors. We introduce Weighted Focal Structure Analysis (WFSA), a method that integrates toxicity scores with structural cohesion. The WFSA identifies focal toxic structures (FTSs), which are cohesive subgroups that amplify harmful content beyond the capabilities of individual users. The method uses micro-level user selection, meso-level community detection, and composite NDCG ranking. We evaluated WFSA on large-scale datasets, including Russian Telegram channels (324,769 users), Reddit Climate Change discussions (1.2 million entries), Twitter (102,913 toxic COVID-19 tweets), and synthetic networks. Monte Carlo epidemic simulations were performed with 1,500 iterations across SIR, SEIR, and SEIZ models to achieve statistical convergence. WFSA demonstrated network-dependent performance, with high-clustering networks achieving an F1 score of 0.92, while low-clustering networks reached a score of 0.56. FTSs exhibited 58.6% higher toxicity propagation compared to individual approaches ( \(p<0.001\) ). SEIZ-MC had superior prediction accuracy, with an error rate of 5.7% compared to 32.9% for SIR-MC. Targeting coordinated groups for moderation proved more effective at containing toxicity than focusing on individuals. This work advances structure-aware anomaly detection in complex social networks, with applications in identifying coordinated manipulation, misinformation campaigns, and analyzing collective behavior on digital platforms.