Processed gene and clinical data

The processed cancer dataset mentioned in the paper "Cox-Sage: Enhancing Cox proportional hazards model with interpretable graph neural networks for cancer prognosis," which is currently under review in Briefings in Bioinformatics. The gene expression data and clinical data of seven cancer types downloaded from TCGA (https://portal.gdc.cancer.gov/) were processed to retain only protein-coding genes. A patient similarity graph was constructed based on the similarity of clinical data. The data for each type of cancer consists of a `gene_expression.csv`, a `clinical.csv`, and an `adj_list.pkl`. In addition, the `prognostic_genes.zip` file contains the hazards contour plot of all prognostic genes identified in the study. And the `all_benchmarks_prediction_results.zip` file contains the hazards prediction results of all benchmarks that being reproduced.本数据集为目前正在《生物信息学简报》（Briefings in Bioinformatics）审稿中的论文《Cox-Sage：利用可解释图神经网络（graph neural networks）增强Cox比例风险模型（Cox proportional hazards model）以实现癌症预后分析》中提及的经预处理的癌症数据集。该数据集从TCGA（https://portal.gdc.cancer.gov/）下载了7种癌症类型的基因表达数据与临床数据，经预处理仅保留蛋白编码基因。基于临床数据的相似性构建了患者相似性图。每一类癌症的数据均包含`gene_expression.csv`、`clinical.csv`与`adj_list.pkl`三个文件。此外，`prognostic_genes.zip`文件包含本研究中鉴定出的全部预后基因的风险等高线图；`all_benchmarks_prediction_results.zip`文件则包含所有已复现基准模型的风险预测结果。