Multidimensional Normalization to Minimize Plate Effects of Suspension Bead Array Data
Multidimensional Normalization to Minimize Plate Effects of Suspension Bead Array Data
复制标题
DOI：
10.1021/acs.jproteome.5b01131
复制
发表时间：
2016-10-01
影响因子：
4.4
通讯作者：
Schwenk, Jochen M.
中科院分区：
文献类型：
--
作者：
Hong, Mun-Gwan;Lee, Woojoo;Schwenk, Jochen M.
随着生物库数量的增加，生物标记物研究现在可以通过使用大样本集以合理的统计能力进行。以抗体为基础的蛋白质组学通过悬浮珠阵列提供了一种吸引人的方法来分析血清、血浆或脑脊液样本，用于在微滴定板上进行这类研究。为了将测量范围扩大到单批次以外，即每片96或384个样品，需要适当的归一化方法以最大限度地减少板之间的差异。这里，我们提出了两种利用MA坐标的归一化方法。多维MA(多MA)和MA-LOISE都考虑了每个悬浮珠阵列分析的微滴定板的所有样品，因此不需要任何外部参考样品。通过对384个样本(包括血清和血浆)的分析，我们展示了两种MA归一化方法的性能。样本在96孔板上随机排列，分别在分析板中进行处理和分析。使用主成分分析(PCA)，我们可以表明，与其他归一化方法相比，多MA归一化方法消除了前两个成分中存在的板状簇。此外，我们研究了随机抗体对之间的相关性，发现两种MA归一化方法都大大减少了由平板效应引入的夸大相关性。使用多MA和MA-LOISE的归一化方法最大限度地减少了用抗体悬浮珠阵列分析几个化验板所产生的批次效应。在一项模拟的生物标记物研究中，多MA恢复了由于板块效应而失去的关联。我们的归一化方法可作为R包MDimNornin，在使用其他类型的高通量分析数据的研究中也可能有用。
Enhanced by the growing number of biobanks, biomarker studies can now be performed with reasonable statistical power by using large sets of samples. Antibody-based proteomics by means of suspension bead arrays offers one attractive approach to analyze serum, plasma, or CSF samples for such studies in microtiter plates. To expand measurements beyond single batches, with either 96 or 384 samples per plate, suitable normalization methods are required to minimize the variation between plates. Here we propose two normalization approaches utilizing MA coordinates. The multidimensional MA (multi-MA) and MA-loess both consider all samples of a microtiter plate per suspension bead array assay and thus do not require any external reference samples. We demonstrate the performance of the two MA normalization methods with data obtained from the analysis of 384 samples including both serum and plasma. Samples were randomized across 96-well sample plates, processed, and analyzed in assay plates, respectively. Using principal component analysis (PCA), we could show that plate-wise clusters found in the first two components were eliminated by multi-MA normalization as compared with other normalization methods. Furthermore, we studied the correlation profiles between random pairs of antibodies and found that both MA normalization methods substantially reduced the inflated correlation introduced by plate effects. Normalization approaches using multi-MA and MA-loess minimized batch effects arising from the analysis of several assay plates with antibody suspension bead arrays. In a simulated biomarker study, multi-MA restored associations lost due to plate effects. Our normalization approaches, which are available as R package MDimNornin, could also be useful in studies using other types of high-throughput assay data.