Physiology Weihenstephan, ZIEL Research Center for Nutrition and Food Sciences, Technische Universität München, Freising, Germany.
PLoS One. 2012;7(6):e38946. doi: 10.1371/journal.pone.0038946. Epub 2012 Jun 18.
Adequate normalization minimizes the effects of systematic technical variations and is a prerequisite for getting meaningful biological changes. However, there is inconsistency about miRNA normalization performances and recommendations. Thus, we investigated the impact of seven different normalization methods (reference gene index, global geometric mean, quantile, invariant selection, loess, loessM, and generalized procrustes analysis) on intra- and inter-platform performance of two distinct and commonly used miRNA profiling platforms.
METHODOLOGY/PRINCIPAL FINDINGS: We included data from miRNA profiling analyses derived from a hybridization-based platform (Agilent Technologies) and an RT-qPCR platform (Applied Biosystems). Furthermore, we validated a subset of miRNAs by individual RT-qPCR assays. Our analyses incorporated data from the effect of differentiation and tumor necrosis factor alpha treatment on primary human skeletal muscle cells and a murine skeletal muscle cell line. Distinct normalization methods differed in their impact on (i) standard deviations, (ii) the area under the receiver operating characteristic (ROC) curve, (iii) the similarity of differential expression. Loess, loessM, and quantile analysis were most effective in minimizing standard deviations on the Agilent and TLDA platform. Moreover, loess, loessM, invariant selection and generalized procrustes analysis increased the area under the ROC curve, a measure for the statistical performance of a test. The Jaccard index revealed that inter-platform concordance of differential expression tended to be increased by loess, loessM, quantile, and GPA normalization of AGL and TLDA data as well as RGI normalization of TLDA data.
CONCLUSIONS/SIGNIFICANCE: We recommend the application of loess, or loessM, and GPA normalization for miRNA Agilent arrays and qPCR cards as these normalization approaches showed to (i) effectively reduce standard deviations, (ii) increase sensitivity and accuracy of differential miRNA expression detection as well as (iii) increase inter-platform concordance. Results showed the successful adoption of loessM and generalized procrustes analysis to one-color miRNA profiling experiments.
充分的归一化可以最小化系统技术差异的影响，是获得有意义的生物学变化的前提。然而，miRNA 归一化性能和建议并不一致。因此，我们研究了七种不同归一化方法（参考基因指数、全局几何平均值、分位数、不变选择、局部线性回归、局部线性回归 M 和广义普罗克鲁斯分析）对两种不同且常用的 miRNA 分析平台的内平台和间平台性能的影响。
方法/主要发现：我们包括了基于杂交的平台（安捷伦科技）和 RT-qPCR 平台（应用生物系统公司）的 miRNA 分析数据。此外，我们通过个体 RT-qPCR 检测验证了一小部分 miRNA。我们的分析纳入了分化和肿瘤坏死因子α处理对原代人骨骼肌细胞和鼠骨骼肌细胞系的影响的数据。不同的归一化方法在以下方面存在差异：（i）标准偏差，（ii）接收器操作特性（ROC）曲线下的面积，（iii）差异表达的相似性。在 Agilent 和 TLDA 平台上，局部线性回归、局部线性回归 M 和分位数分析在最小化标准偏差方面最为有效。此外，局部线性回归、局部线性回归 M、不变选择和广义普罗克鲁斯分析增加了 ROC 曲线下的面积，这是测试统计性能的一个指标。Jaccard 指数表明，通过对 AGL 和 TLDA 数据进行 loess、loessM、quantile 和 GPA 归一化以及对 TLDA 数据进行 RGI 归一化，可以提高平台间差异表达的一致性。
结论/意义：我们建议对 miRNA Agilent 阵列和 qPCR 卡应用局部线性回归或局部线性回归 M 和 GPA 归一化，因为这些归一化方法可以：（i）有效地降低标准偏差，（ii）提高差异 miRNA 表达检测的敏感性和准确性，以及（iii）提高平台间的一致性。结果表明，局部线性回归 M 和广义普罗克鲁斯分析可以成功应用于双色 miRNA 分析实验。
