Vanderbilt Memory and Alzheimer's Center, Vanderbilt University Medical Center, Nashville, TN 37212, USA.
Pac Symp Biocomput. 2022;27:361-372.
Gene-based methods such as PrediXcan use expression quantitative trait loci to build tissue-specific gene expression models when only genetic data is available. There are known sex differences in tissue-specific gene expression and in the genetic architecture of gene expression, but such differences have not been incorporated into predicted gene expression models to date. We built sex-aware PrediXcan models using whole blood transcriptomic data from the Genotype-Tissue Expression (GTEx) project (195 females and 371 males) and evaluated their performance in an independent dataset. Specifically, PrediXcan models were built following the method described in Gamazon et al. 2015, but we included both whole-sample and sex-specific models. Validation was evaluated leveraging lymphoblast RNA sequencing data from the EUR cohort of the 1000 Genomes Project (178 females and 171 males). Correlations (R2) between observed and predicted expression were evaluated in 5,283 autosomal genes to determine performance of models. In sum, we successfully predicted 1,149 genes in males and 623 in females, while 3,511 genes appeared to be not sex-specific. Of the sex-specific genes, 15% (189 genes in males and 73 genes in females) exhibited higher R2 in sex-specific models compared to whole-sample models, although the overall gain in predictive power was generally minimal and well within measurement error. Nevertheless, two female-specific genes and six male-specific genes showed significantly better prediction when using the sex-specific weights versus the whole-sample weights; furthermore, several of these genes play a role in mitochondrial metabolism, which is known to be influenced by sex hormones. Taken together, these results support previous reports of the small contribution of genetic architecture to sex-specific expression. Still, sex-aware PrediXcan models were able to provide robust sex-specific prediction signals. Future studies exploring the contribution of the X chromosome and tissue specificity on sex-specific genetically regulated expression will clarify the utility of this method.
基于基因的方法，如 PrediXcan，当只有遗传数据可用时，可以使用表达数量性状基因座来构建组织特异性基因表达模型。组织特异性基因表达和基因表达的遗传结构中存在已知的性别差异，但迄今为止，这些差异尚未纳入预测基因表达模型中。我们使用来自基因型组织表达 (GTEx) 项目的全血转录组数据 (195 名女性和 371 名男性) 构建了具有性别意识的 PrediXcan 模型，并在独立数据集上评估了它们的性能。具体来说，PrediXcan 模型是按照 Gamazon 等人在 2015 年描述的方法构建的，但我们包括了全样本和性别特异性模型。验证是通过利用来自 1000 个基因组项目 EUR 队列的淋巴母细胞 RNA 测序数据来评估的。在 5283 个常染色体基因中评估了观察到的和预测到的表达之间的相关性 (R2)，以确定模型的性能。总的来说，我们成功地预测了 1149 个男性基因和 623 个女性基因，而 3511 个基因似乎没有性别特异性。在性别特异性基因中，15%（男性的 189 个基因和女性的 73 个基因）在性别特异性模型中表现出比全样本模型更高的 R2，尽管整体预测能力的提高通常很小，并且很好地符合测量误差。尽管如此，当使用性别特异性权重而不是全样本权重时，两个女性特异性基因和六个男性特异性基因的预测表现明显更好；此外，这些基因中的几个在已知受性激素影响的线粒体代谢中发挥作用。总的来说，这些结果支持先前关于遗传结构对性别特异性表达贡献较小的报告。尽管如此，具有性别意识的 PrediXcan 模型仍然能够提供稳健的性别特异性预测信号。未来探索 X 染色体和组织特异性对性别特异性遗传调控表达的贡献的研究将阐明该方法的实用性。
