Abstract: The cell graph data extracted from histological images for predicting microstatellite status. This is a patch-level (patch refers to the sub-image tessellated from the whole-slide image) binary classification task: the two datasets classify images patches to either MSS (microsatellite stable) or MSIMUT (microsatellite instable or highly mutated). The original histology patches are available from [1]. Cell Graph Contraction: In this dataset, we establish a cell graph for each patch. The cell graph can represent the cell-cell interaction and the collection of cell graphs for all patches provide a precise characterization of tumor microenvironment. With only the availability of raw image patches, we leverage the nuclei regions segmented by a well-tuned CA2.5-Net [2] to extract the node features of each single nuclei node. And then we extract a total number of 94 pre-defined pathomics features for each nuclei region as the corresponding graph node feature. As the morphological signals are believed relative to cell-cell interplay, the cell-specific features which include the nuclei coordination, optical, and representations, then characterize the cell-level morphological behavior. We then calculate the pair-wise Euclidean distance between nuclei centroids to establish edges of a cell graph Reference For more details about the construction of cell-graph, please visit our paper: https://arxiv.org/abs/2206.07599. The associated codes and usage for the datasets are available at: https://github.com/yiqings/HEGnnEnhanceCnn. [1] Histological images for MSI vs. MSS classification in gastrointestinal cancer, FFPE samples. https://doi.org/10.5281/zenodo.2530835 [2] Huang, J., Shen, Y., Shen, D., & Ke, J. (2021, September). CA 2.5-Net Nuclei Segmentation Framework with a Microscopy Cell Benchmark Collection. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 445-454). Springer, Cham.