Wang Xi, Gardiner Erin J, Cairns Murray J
School of Biomedical Sciences and Pharmacy, Faculty of Health and Medicine, The University of Newcastle, NSW 2308, Australia.
Mol Biosyst. 2015 May;11(5):1235-40. doi: 10.1039/c4mb00711e.
Normalization of high-throughput molecular expression profiles secures differential expression analysis between samples of different phenotypes or biological conditions, and facilitates comparison between experimental batches. While the same general principles apply to microRNA (miRNA) normalization, there is mounting evidence that global shifts in their expression patterns occur in specific circumstances, which pose a challenge for normalizing miRNA expression data. As an alternative to global normalization, which has the propensity to flatten large trends, normalization against constitutively expressed reference genes presents an advantage through their relative independence. Here we investigated the performance of reference-gene-based (RGB) normalization for differential miRNA expression analysis of microarray expression data, and compared the results with other normalization methods, including: quantile, variance stabilization, robust spline, simple scaling, rank invariant, and Loess regression. The comparative analyses were executed using miRNA expression in tissue samples derived from subjects with schizophrenia and non-psychiatric controls. We proposed a consistency criterion for evaluating methods by examining the overlapping of differentially expressed miRNAs detected using different partitions of the whole data. Based on this criterion, we found that RGB normalization generally outperformed global normalization methods. Thus we recommend the application of RGB normalization for miRNA expression data sets, and believe that this will yield a more consistent and useful readout of differentially expressed miRNAs, particularly in biological conditions characterized by large shifts in miRNA expression.
高通量分子表达谱的标准化可确保对不同表型或生物学条件的样本进行差异表达分析，并便于实验批次之间的比较。虽然相同的一般原则适用于微小RNA（miRNA）的标准化，但越来越多的证据表明，在特定情况下会出现其表达模式的整体变化，这给miRNA表达数据的标准化带来了挑战。作为全局标准化的替代方法（全局标准化倾向于使大趋势变平），针对组成型表达的参考基因进行标准化具有相对独立性的优势。在这里，我们研究了基于参考基因（RGB）的标准化在微阵列表达数据差异miRNA表达分析中的性能，并将结果与其他标准化方法进行了比较，包括：分位数、方差稳定化、稳健样条、简单缩放、秩不变和局部加权回归。使用来自精神分裂症患者和非精神疾病对照受试者的组织样本中的miRNA表达进行了比较分析。我们提出了一个一致性标准，通过检查使用整个数据的不同分区检测到的差异表达miRNA的重叠来评估方法。基于这个标准，我们发现RGB标准化通常优于全局标准化方法。因此，我们建议将RGB标准化应用于miRNA表达数据集，并相信这将产生更一致和有用的差异表达miRNA读数，特别是在以miRNA表达大幅变化为特征的生物学条件下。
