Hein Anne-Mette K, Richardson Sylvia
Dept. of Epidemiology and Public Health, Imperial College London, Norfolk Place, London, UK.
BMC Bioinformatics. 2006 Jul 20;7:353. doi: 10.1186/1471-2105-7-353.
Studies of differential expression that use Affymetrix GeneChip arrays are often carried out with a limited number of replicates. Reasons for this include financial considerations and limits on the available amount of RNA for sample preparation. In addition, failed hybridizations are not uncommon leading to a further reduction in the number of replicates available for analysis. Most existing methods for studying differential expression rely on the availability of replicates and the demand for alternative methods that require few or no replicates is high.
We describe a statistical procedure for performing differential expression analysis without replicates. The procedure relies on a Bayesian integrated approach (BGX) to the analysis of Affymetrix GeneChips. The BGX method estimates a posterior distribution of expression for each gene and condition, from a simultaneous consideration of the available probe intensities representing the gene in a condition. Importantly, posterior distributions of expression are obtained regardless of the number of replicates available. We exploit these posterior distributions to create ranked gene lists that take into account the estimated expression difference as well as its associated uncertainty. We estimate the proportion of non-differentially expressed genes empirically, allowing an informed choice of cut-off for the ranked gene list, adapting an approach proposed by Efron. We assess the performance of the method, and compare it to those of other methods, on publicly available spike-in data sets, as well as in a proper biological setting.
The method presented is a powerful tool for extracting information on differential expression from GeneChip expression studies with limited or no replicates.
使用Affymetrix基因芯片阵列进行的差异表达研究通常在复制品数量有限的情况下进行。这样做的原因包括财务考虑以及样本制备中可用RNA量的限制。此外，杂交失败并不罕见，这导致可供分析的复制品数量进一步减少。大多数现有的研究差异表达的方法依赖于复制品的可用性，因此对几乎不需要或不需要复制品的替代方法的需求很高。
我们描述了一种无需复制品即可进行差异表达分析的统计程序。该程序依赖于一种贝叶斯综合方法（BGX）来分析Affymetrix基因芯片。BGX方法通过同时考虑代表某一条件下某一基因的可用探针强度，估计每个基因和条件下表达的后验分布。重要的是，无论可用复制品的数量如何，都能获得表达的后验分布。我们利用这些后验分布来创建排名基因列表，该列表考虑了估计的表达差异及其相关的不确定性。我们根据经验估计非差异表达基因的比例，从而能够明智地选择排名基因列表的截止值，这是采用了Efron提出的一种方法。我们在公开可用的掺入数据集中以及在适当的生物学环境中评估了该方法的性能，并将其与其他方法的性能进行了比较。
所提出的方法是一种强大的工具，可用于从有限或无复制品的基因芯片表达研究中提取差异表达信息。
