CONTENT -- Multi-context genetic modeling TWAS and eAssociation summary statistics

We provide the summary statistics of running CONTENT, the context-by-context approach, and UTMOST on over 22 phenotypes. The phenotypes are listed in the manuscript, and their respective studies and sample size can be found in a table under the supplementary section of the manuscript. All 3 methods were trained on GTEx v7 as well as CLUES, a single-cell RNA sequencing dataset of PBMCs. The data include the gene name, model, cross-validated R^2, prediction pvalue, TWAS p value, TWAS Z score, and a column titled "hFDR" indicating whether the association was statistically significant while employing hierarchical FDR. The benefits of employing such an approach for all methods can be found in the manuscript. We also include the eAssociations that we obtain by training prediction models on GTEx and CLUES alone. For the CxC and UTMOST approaches, these files contain the gene, context, pvalue and adjusted R^2. For CONTENT, these include the gene, context and pvalue and adjusted R^2 for each CONTENT model--the column names are described like a regression of y~x, rsq_y_x, so rsq_observed_full is the adjusted R^2 from regressing the observed expression onto the cross-validated full model predictions. In cases where the R^2 is higher from the specific or shared models, it's best to use either of those rather than the full model for out of sample prediction.

本数据集提供了CONTENT、逐上下文分析方法（context-by-context approach，简称CxC）以及UTMOST三种方法在超过22种表型上的运行汇总统计量。手稿中列出了所有表型，各表型对应的研究及样本量可参见手稿补充材料部分的附表。三种方法均基于基因型-组织表达（GTEx）v7数据集以及CLUES（外周血单个核细胞（PBMC）的单细胞RNA测序数据集）完成训练。数据集包含以下字段：基因名称、模型、交叉验证决定系数（R²）、预测p值、转录组全关联分析（Transcriptome-Wide Association Study，TWAS）p值、TWAS Z得分，以及一列名为"hFDR"的指标，用于说明采用分层错误发现率（hierarchical FDR，hFDR）时，该关联是否具有统计学显著性。三种方法采用此类分析策略的优势可参见本手稿。本数据集还包含仅基于GTEx与CLUES训练预测模型所得到的表达关联（eAssociations）。针对逐上下文分析方法与UTMOST方法，对应文件包含基因、上下文、p值及校正后决定系数（adjusted R²）。针对CONTENT方法，对应文件则包含各CONTENT模型对应的基因、上下文、p值及校正后决定系数——其列名采用y~x形式的回归命名范式，以rsq_y_x作为格式示例，其中rsq_observed_full指将观测到的基因表达量对交叉验证后的全模型预测值进行回归所得到的校正后决定系数。若特定模型或共享模型的决定系数（R²）更高，则在样本外预测中优先选用此类模型而非全模型。