On the pros and cons of mega-analyses compared to meta-analyses for genome-wide association studies

全基因组关联研究在确定与复杂疾病相关的遗传变异方面非常成功。为了增加样本量和统计能力，这些研究通常以荟萃分析的方式进行，结合研究特定的汇总统计数据，对50万至3000万个基因分型和输入的变异进行统计。据推测，收集个体参与者数据（IPD）以在一个非常大的数据集（大型分析）中实现联合质量控制、联合imputation和联合建模，而不是对汇总统计数据进行荟萃分析，将提高检测疾病位点的能力。然而，很少有研究评估大型分析与荟萃分析的哪个方面有助于在真实数据场景中获得多少数据质量和疾病位点检测。由于基因型输入的高计算负担和缺乏大的全基因组IPD数据，这种评估仍然有限。问题仍然是，大型分析能带来多少好处，以及这种努力是否值得。因此，我们将着手对大型IPD的大型分析的各个方面进行系统评估，而不是对特定研究汇总统计数据进行荟萃分析。我们将量化数据质量的提高和在真正的大型全基因组IPD中检测疾病位点的能力，并将来自真实数据集的信息与模拟方法相结合，以扩展场景。我们将从三个层面进行比较，即质量控制、imputation和建模。一个特别的焦点将是罕见的变体。根据我们以前的工作，我们处于一个独特的情况下，我们手头有一个大的IPD （bb40000人患有和没有年龄相关性黄斑变性）。这种疾病和我们的数据提供了一个理想的榜样，因为数据集由1,120万个遗传变异组成，其中包括16万个罕见的蛋白质改变，并显示出34个可检测的AMD位点。该项目的结果将指导未来的遗传研究，是否值得为收集IPD数据进行大规模分析而付出努力，并改进疾病位点检测。该项目的结果还将有助于了解如何分析所有数据都可作为IPD提供的大型多中心GWAS。

Genome-wide association studies have been highly successful in identifying genetic variants associated with complex diseases. To increase the sample size and statistical power, these studies are usually conducted as meta-analyses combining study-specific aggregated statistics for each of 0.5 to 30 million genotyped and imputed variants. It is hypothesized that collecting individual participant data (IPD) to enable joint quality control, joint imputation, and joint modelling in one very big data set (mega-analysis) instead of a meta-analysis of aggregated statistics would improve the ability to detect disease loci. However, there is little work evaluating what aspect of mega-analysis versus meta-analysis contributes to how much gain in data quality and disease loci detection in real data scenarios. Such evaluations are still limited due to the high computational burden of genotype imputing and the lack of large genome-wide IPD data. The question remains how much can be gained by mega-analysis and whether it is worth the effort. We thus will set out to conduct a systematic evaluation of each aspect of a mega-analysis of large IPD versus a meta-analysis of study-specific aggregated statistics. We will quantify the gain in the data quality and the ability to detect disease loci in a real large genome-wide IPD and will combine information from the real data set with simulation approaches to expand the scenarios. We will conduct the comparisons on three levels, the quality control, the imputation, and the modelling. One special focus will be rare variants.Based on our previous work, we are in the unique situation of having a large IPD at our hand (> 40.000 persons with and without age-related macular degeneration). This disease and our data provide an ideal role model, since the data set consists of > 12 million genetic variants including 160,000 that are rare protein-altering, and exhibits 34 detectable AMD loci. The results of this project will guide future genetic studies whether the challenge to gather IPD data for mega-analysis will be worth the effort and improve disease locus detection. The results of this project will also help understand how to analyse large multi-center GWAS where all data is available as IPD.