Examples of coexpressions from literature.

MotivationCoexpression estimations are helpful for analysis of pathways, cofactors, regulators, targets, and human health and disease. Ideally, coexpression estimations should consider as many diverse cell types as possible and consider that available data is not uniform across tissues. Importantly, the coexpression estimations accessible today are performed on a “tissue level”, which is based on cell type standardized formulations. Little or no attention is paid to overall gene expression levels. The tissue-level estimation assumes that variance expression levels are more important than mean expression levels. Here, we challenge this assumption by estimating a coexpression calculation at the “system level”, which is estimated without standardization by tissue, and show that it provides valuable information. We made available a resource to view, download, and analyze both, tissue- and system-level coexpression estimations from GTEx human data.MethodsGTEx v8 expression data was globally normalized, batch-processed, and filtered. Then, PCA, clustering, and tSNE stringent procedures were applied to generate 42 distinct and curated tissue clusters. Coexpression was estimated from these 42 tissue clusters computing the correlation of 33,445 genes by sampling 70 samples per tissue cluster to avoid tissue overrepresentation. This process was repeated 20 times, extracting the minimum value provided as a robust estimation. Three metrics were calculated (Pearson, Spearman, and G-statistic) in two data processing modes, at the system-level (TPM scale) and tissue levels (z-score scale).ResultsWe first validate our tissue-level estimations compared with other databases. Then, by specific analyses in several examples and literature validations of predictions, we show that system-level coexpression estimation differs from tissue-level estimations and that both contain valuable information reflected in biological pathways. We also show that coexpression estimations are associated to transcriptional regulation. Finally, we present CoGTEx, a valuable resource for viewing and analyzing coexpressed genes in human adult tissues from GTEx v8 data. We introduce our web resource to list, view and explore the coexpressed genes from GTEx data.ConclusionWe conclude that system-level coexpression is a novel and interesting coexpression metric capable of generating plausible predictions and biological hypotheses; and that CoGTEx is a valuable resource to view, compare, and download system- and tissue- level coexpression estimations from GTEx data.AvailabilityThe web resource is available at http://bioinformatics.mx/cogtex.研究背景：共表达（coexpression）估计可为通路分析、辅因子研究、调控因子解析、靶标筛选以及人类健康与疾病相关研究提供有力支撑。理想状态下，共表达估计应尽可能涵盖多样的细胞类型，同时需考虑不同组织的可用数据并不均衡。值得关注的是，当前可获取的共表达估计均采用“组织层面”分析范式，该范式基于细胞类型标准化方案开展。此类估计极少甚至未考虑基因的整体表达水平，且默认表达方差相较于均值更为重要。本研究对这一假设提出挑战：我们通过“系统层面”开展共表达计算——该计算未按组织进行标准化——并证实其可提供极具价值的生物学信息。我们公开了一款资源平台，可用于查看、下载并分析来自GTEx（Genotype-Tissue Expression）人类数据集的组织层面与系统层面共表达估计结果。 研究方法：本研究对GTEx v8版的表达数据进行了全局标准化、批次校正与过滤处理。随后通过严格的主成分分析（Principal Component Analysis, PCA）、聚类及t分布邻域嵌入（t-distributed Stochastic Neighbor Embedding, tSNE）流程，得到42个独立且经过精细注释的组织簇。基于这42个组织簇，我们对33445个基因的相关性进行计算以估计共表达：每个组织簇抽取70个样本以避免组织占比过高带来的偏差。该流程重复20次，取最小值作为稳健估计结果。我们在两种数据处理模式下计算了三种评估指标：皮尔逊相关系数、斯皮尔曼相关系数与G统计量（G-statistic），其中系统层面采用每百万转录本（Transcripts Per Million, TPM）标度，组织层面采用z分数（z-score）标度。 研究结果：我们首先将组织层面的共表达估计结果与其他数据库进行比对验证。随后通过多个实例的针对性分析以及对预测结果的文献验证，证实系统层面的共表达估计与组织层面的估计存在显著差异，且二者均包含反映生物学通路的有价值信息。我们还发现共表达估计与转录调控存在关联。最后，我们推出了CoGTEx资源平台，可用于查看、分析来自GTEx v8版数据的人类成人组织共表达基因。本研究还介绍了该网页资源，可用于检索、查看与探索GTEx数据中的共表达基因。 研究结论：本研究证实，系统层面的共表达是一种全新且极具研究价值的共表达评估指标，可生成合理的预测结果与生物学假说；同时，CoGTEx是一款极具价值的资源平台，可用于查看、对比并下载来自GTEx数据的系统层面与组织层面共表达估计结果。 数据可用性：本网页资源的访问地址为 http://bioinformatics.mx/cogtex。