Histopathological images employ both pixel-based and graph-based methods to capture microenvironmental patterns, significantly enhancing machine learning applications for various downstream tasks. However, their substantial size poses considerable computational challenges due to gigapixel images and graphs with thousands of nodes. Traditional localized analysis techniques, such as patch-based methods, assume uniform labels and fail to provide a comprehensive understanding of biological entities and their context. In this work, we propose the use of graph coarsening techniques for the compression of cell graphs, addressing computational challenges while preserving critical information. Our approach reduces graph size by maintaining cellular morphology, topology, and spatial relationships, thereby preserving interpretability across diverse applications. This enables analysis at the whole slide image (WSI) level using cell graph representations, avoiding the limitations of patch-based methods. We validate this approach on the breast cancer dataset for multi-class cancer subtyping, demonstrating that our coarsened graphs achieve performance comparable to state-of-the-art models. Furthermore, the preserved explainability of our method confirms the retention of essential information in the coarsened graphs. Our coarsened graphs surpass both image-based and original cell graph representations in computational efficiency and storage, advancing histopathological image analysis for various downstream tasks.