In order to accommodate myriad disparate services, from legacy voice and data, to niche network applications for industry verticals, 6G networks are expected to heavily exploit the concept of network slicing introduced in 5G. However, the increased complexity of sliced networks amplifies the risk of configuration errors, necessitating the expanded use of Network Digital Twins (NDTs) for proactive service assurance. Legacy NDTs, are ill-suited to highly virtualized environments, as they fail to model the impact of virtualization on physical infrastructure and overlook long-term dependency effects. In this work, we highlight the performance impact of virtualization and introduce an attention-enhanced Graph Neural Networks (GNN)-based NDT to address these challenges. Our simulation results demonstrate that the proposed NDT framework significantly outperforms state-of-the-art models in accurately predicting service Key Performance Indicators (KPIs) across disparate use cases.