ABSTRACT

Recent years have seen a surge of methylome-wide association studies (MWAS). We observed that many of these studies suffer from test statistic inflation that is most likely caused by commonly used quality control (QC) pipelines not going far enough to remove technical artefacts. To support this claim, we reanalysed GEO datasets with an improved QC pipeline that reduced test-statistic inflation parameter lambda from the original mean/median of 20.16/15.17 to 3.07/1.14. Furthermore, the mean/median number of methylome-wide significant findings was reduced by 65,688/57,805 loci after more thorough QC. To avoid such false positives we argue for more extensive QC and that reporting the test-statistic inflation parameter lambda become standard for all MWAS allowing readers to better assess the risk of false discoveries.

摘要

近年来，甲基化组范围内的关联研究 (MWAS) 激增。我们观察到，其中许多研究都存在测试统计膨胀的问题，这很可能是由于常用的质量控制 (QC) 管道不足以消除技术人工制品造成的。为了支持这一说法，我们使用改进的 QC 管道重新分析了 GEO 数据集，该管道将测试统计膨胀参数 lambda 从原来的平均值/中位数 20.16/15.17 降低到 3.07/1.14。此外，在更彻底的 QC 后，甲基化组范围内显着发现的平均/中位数减少了 65,688/57,805 个位点。为了避免此类误报，我们主张进行更广泛的 QC，并且报告测试统计膨胀参数 lambda 成为所有 MWAS 的标准，使读者能够更好地评估错误发现的风险。