【佳学基因检测】精神病基因组协会如何解码疾病发生的基因原因并应用基因检测？

什么是影像表型？

利用英国生物银行的大规模图像数据和精神病基因组协会的大规模GWAS数据的方法有可能开启对精神疾病生物学的许多洞察。在本文中，我们提出了一种这样的方法，BrainXcan，它利用这两种数据资源来解决小规模MRI研究中的一些不足。以英国生物银行的数据为参考，我们建立了从基因数据预测大脑IDPs的模型。这些模型可以应用于全基因组关联研究。例如，使用精神病基因组协会收集的精神分裂症GWAS数据，我们的方法测试了精神分裂症与许多不同功能、结构和扩散MR模式之间的关联，大小为∼ 70000个案例和∼ 24万个控件。此外，通过应用孟德尔随机方法，我们推断出因果关系的方向：IDP的变化是疾病的原因还是后果。
Methods that leverage UK Biobank’s large scale image data and the PGC’s large scale GWAS data have the potential to unlock many insights into the biology of mental disorders. In this paper we propose one such method, BrainXcan, which leverages these two data resources to address some of the deficiencies in small scale MRI studies. Using UK Biobank data as a reference, we build models to predict brain IDPs from genetic data. These models can then be applied to from genome-wide association studies. For example, using the schizophrenia GWAS data collected by the PGC, our method tests for association between schizophrenia and a number of different functional, structural and diffusion MR modalities with size of ∼ 70, 000 cases and ∼ 240, 000 controls. Furthermore, by applying a Mendelian randomization approach we infer the direction of causality: whether the changes in IDP are the cause of disease or a consequence of it.
影像表型（IDP）相关遗传标记已被用于因果推断，并采用孟德尔随机法等方法，在大样本量和防止反向因果关系的情况下，研究大脑特征对行为表型的中介作用。例如，Jansen等人（2020年）研究了脑容量IDP和智力之间共享的基因组位点和相应基因，他们确定了92个共享基因，为脑容量和智力的共享遗传病因学提供了见解。Shen等人（2020年）对抑郁症和dMRI IDPs进行了双向MR分析，发现提示性证据表明丘脑辐射平均扩散率的变化可能是抑郁症的后果。一种相关的方法是将遗传预测的大脑IDP/表型与复杂性状相关联，这是基于转录组的方法（Gamazon等人，2015；Gusev等人，2016）对IDP的延伸。基于这一想法，Knutson等人（2020年）利用阿尔茨海默病神经成像倡议的14个大脑特征开展了成像广泛关联研究（IWAS）。他们还使用标准PRS方法，使用Elliott等人（2018年）（n=8428）的GWAS汇总结果生成预测权重。
IDP-associated genetic markers have been used for causal inference with methods such as Mendelian Randomization to investigate the mediating role of brain features on behavioral phenotypes with both large sample sizes and protection from reverse causality. For instance, studied the genomic loci and corresponding genes that are shared between brain volume IDPs and intelligence and they identified 92 shared genes which provided insight of the shared genetic etiology of brain volume and intelligence. Shen et al. (2020) performed bi-directional MR analysis with depression and dMRI IDPs finding suggestive evidence that the change of the mean diffusivity in thalamic radiations could be a consequence of major depressive disorder. A related approach is one that correlates genetically predicted brain IDP/phenotype and the complex trait, an extension of transcriptome-based methods (Gamazon et al., 2015; ) to IDPs. Based on this idea, developed imaging-wide association study (IWAS) using 14 brain features from the Alzheimer’s Disease Neuroimaging Initiative. They also used standard PRS approaches to generate prediction weights using the GWAS summary results from Elliott et al. (2018) (n=8,428).

在本文中，我们对IDPs的遗传结构进行了深入分析，并进一步处理英国生物银行的IDPs，以开发一个最大化可解释性、鲁棒性、计算效率和用户友好性的框架。

大脑特征的高度多基因性对现有方法提出了若干挑战，限制了检测其与疾病的联系的能力；孟德尔随机化方法所需的强大遗传工具很难识别。我们通过开发IDPs的多基因预测因子来应对这些挑战，这些预测因子由IDPs的复杂遗传结构和相关结构提供信息。为了便于解释结果，我们开发了区域特异性和全脑预测因子，提供了对潜在偏差的深入分析和量化。我们确保该实现在计算效率上是有效的，并且可扩展到全基因组生物库规模的数据。我们开发了关联方法的扩展，该方法可以使用日益可用的GWAS汇总结果推断关联，即不需要使用单个级别的数据。我们添加了一个孟德尔随机模型来估计因果流的方向。我们通过将其应用于44种人类特征来说明该方法的威力。最后，我们提供了软件、推荐管道和自动化报告，以提高可用性，降低不太熟悉基因研究的用户采用的障碍。

In this paper, we perform an in-depth analysis of the genetic architecture of IDPs and further process UK Biobank’s IDPs to develop a framework that maximizes interpretability, robustness, computational efficiency, and user friendliness.

The high polygenicity of brain features imposes several challenges to existing methods limiting the power to detect their link to diseases; strong genetic instruments needed for Mendelian randomization based approaches are difficult to identify. We address these challenges by developing polygenic predictors of IDPs informed by their complex genetic architecture and correlation structure. To facilitate interpretation of the results, we develop region-specific and brain-wide predictors providing an in-depth analysis and quantification of potential biases. We make sure that the implementation is computationally efficient and scalable to genome-wide Biobank-scale data. We develop an extension of the association method that can infer the association using the increasingly available GWAS summary results, i.e. without the need to use individual level data. We add a Mendelian Randomization module to estimate the direction of the causal flow. We illustrate the power of the approach by applying it to 44 human traits. Finally, we provide the software, the recommended pipeline, and automated reports to improve usability and lower the barrier to adoption for users less familiar with genetic studies.

(责任编辑：佳学基因)