Wilson disease (WD) is a rare genetic disorder characterized by impaired copper metabolism resulting from mutations in the ATP7B gene. This condition leads to severe hepatic and neurological complications. The rarity of WD, coupled with data privacy concerns, presents significant challenges in its diagnosis, prognosis, and treatment optimization. To address these issues, this study proposes a novel hybrid model that integrates Federated Learning (FL) and Deep Reinforcement Learning (DRL). The FL component, implemented using Federated Averaging (FedAvg), enables privacy-preserving multi-institutional collaboration by training a global predictive model without the need to share raw patient data. Meanwhile, the DRL component, leveraging a Deep Q- Network (DQN), dynamically optimizes personalized treatment strategies by adapting to individual patient responses. The effectiveness of the proposed hybrid model is validated using the Online Mendelian Inheritance in Man (OMIM) dataset as the testbed, showcasing its potential to improve diagnostic accuracy, enhance privacy protection, and personalize treatment for WD. OMIM offers a comprehensive catalog of ATP7B gene mutations and associated clinical phenotypes, serving as a critical resource for feature extraction and model training. Comparative analysis against baseline supervised and unsupervised learning approaches demonstrates the superior performance of the hybrid model in terms of accuracy, adaptability, and patient-specific treatment outcomes. The federated framework ensures data security and promotes ethical multi-institutional data sharing, addressing a major bottleneck in rare disease research. This study provides a scalable, privacy-conscious, and adaptive framework for advancing WD diagnosis and treatment. The integration of FL and DRL represents a transformative approach to leveraging AI in precision medicine for rare genetic disorders.