Complex human diseases are affected by genetic and environmental risk factors and their interactions. Gene–environment interaction (GEI) tests for aggregate genetic variant sets have been developed in recent years. However, existing statistical methods become rate limiting for large biobank‐scale sequencing studies with correlated samples. We propose efficient Mixed‐model Association tests for GEne–Environment interactions (MAGEE), for testing GEI between an aggregate variant set and environmental exposures on quantitative and binary traits in large‐scale sequencing studies with related individuals. Joint tests for the aggregate genetic main effects and GEI effects are also developed. A null generalized linear mixed model adjusting for covariates but without any genetic effects is fit only once in a whole genome GEI analysis, thereby vastly reducing the overall computational burden. Score tests for variant sets are performed as a combination of genetic burden and variance component tests by accounting for the genetic main effects using matrix projections. The computational complexity is dramatically reduced in a whole genome GEI analysis, which makes MAGEE scalable to hundreds of thousands of individuals. We applied MAGEE to the exome sequencing data of 41,144 related individuals from the UK Biobank, and the analysis of 18,970 protein coding genes finished within 10.4 CPU hours.

中文翻译：

---

用于大型生物库规模测序研究的高效基因-环境相互作用测试。

复杂的人类疾病受遗传和环境风险因素及其相互作用的影响。近年来，已经开发了用于聚合遗传变异集的基因-环境相互作用 (GEI) 测试。然而，现有的统计方法成为具有相关样本的大型生物库规模测序研究的速率限制。我们提出了有效的基因-环境相互作用（MAGEE）混合模型关联测试，用于在与相关个体的大规模测序研究中测试聚合变异集和环境暴露之间的定量和二元性状的 GEI。还开发了总体遗传主效应和 GEI 效应的联合测试。调整协变量但没有任何遗传效应的空广义线性混合模型在全基因组 GEI 分析中仅拟合一次，从而大大减少整体计算负担。通过使用矩阵投影考虑遗传主效应，将变异集的分数测试作为遗传负担和方差分量测试的组合进行。在全基因组 GEI 分析中，计算复杂度显着降低，这使得 MAGEE 可扩展到数十万个人。我们将 MAGEE 应用于来自英国生物银行的 41,144 名相关个体的外显子组测序数据，并在 10.4 个 CPU 小时内完成了对 18,970 个蛋白质编码基因的分析。这使得 MAGEE 可扩展到数十万个人。我们将 MAGEE 应用于来自英国生物银行的 41,144 名相关个体的外显子组测序数据，并在 10.4 个 CPU 小时内完成了对 18,970 个蛋白质编码基因的分析。这使得 MAGEE 可扩展到数十万个人。我们将 MAGEE 应用于来自英国生物银行的 41,144 名相关个体的外显子组测序数据，并在 10.4 个 CPU 小时内完成了对 18,970 个蛋白质编码基因的分析。