Nonlinear analyses possess tremendous significance throughout the entire lifespans of civil structures. In recent years, the interest in leveraging deep learning (DL) to address the efficiency limitations of the traditional structural analysis methods has increased. However, full-range nonlinear analyses of different structures remain underresearched because of a lack of appropriate data representations and the failure to consider both internal structural information and external load conditions. A heterogeneous graph (HetG) representation scheme that can digitalize arbitrary structural systems with high fidelity is proposed in this study. Furthermore, a composite feature learning framework is developed to enable efficient full-range nonlinear analyses. This framework comprises two main components: ① a heterogeneous graph neural network (GNN)-based module that encodes static features into embeddings with full structural semantics and ② a sequence-to-sequence (Seq2Seq) module that predicts history-dependent responses using structural embeddings and external stimuli in an end-to-end manner. A computational model named structural analysis based on a graph neural network-nonlinear (StructGNN-N) is implemented based on the proposed methodology and is validated through numerical experiments involving real-world concrete structures. The results show that StructGNN-N successfully reproduces the full-range nonlinear responses of all nodes in the entire structure and exhibits excellent generalizability across structures with diverse topological designs and member configurations. Notably, the developed model achieves a computational efficiency level that is 1000 times greater than that of the traditional elastoplastic history analysis approach using the finite-element (FE) method. A parametric analysis and ablation studies demonstrate the effectiveness of the StructGNN-N architecture. Due to its superior accuracy and computational efficiency, the proposed method holds great potential for use in engineering applications, especially in the context of digital twins. This approach provides an inspiring path for simulating diverse engineering structures with accurate and comprehensive mechanical information in real time.