Papsdorf Katharina, Sima Siyuan, Richter Gerhard, Richter Klaus
Center of integrated protein science at the Technische Universität München, Department Chemie, Lichtenbergstr. 4, 85748 Garching, Germany.
Address:
Microb Cell. 2016 Apr 21;3(6):236-247. doi: 10.15698/mic2016.06.505.
DNA-Microarrays are powerful tools to obtain expression data on the genome-wide scale. We performed microarray experiments to elucidate the transcriptional networks, which are up- or down-regulated in response to the expression of toxic polyglutamine proteins in yeast. Such experiments initially generate hit lists containing differentially expressed genes. To look into transcriptional responses, we constructed networks from these genes. We therefore developed an algorithm, which is capable of dealing with very small numbers of microarrays by clustering the hits based on co-regulatory relationships obtained from the SPELL database. Here, we evaluate this algorithm according to several criteria and further develop its statistical capabilities. Initially, we define how the number of SPELL-derived co-regulated genes and the number of input hits influences the quality of the networks. We then show the ability of our networks to accurately predict further differentially expressed genes. Including these predicted genes into the networks improves the network quality and allows quantifying the predictive strength of the networks based on a newly implemented scoring method. We find that this approach is useful for our own experimental data sets and also for many other data sets which we tested from the SPELL microarray database. Furthermore, the clusters obtained by the described algorithm greatly improve the assignment to biological processes and transcription factors for the individual clusters. Thus, the described clustering approach, which will be available through the ClusterEx web interface, and the evaluation parameters derived from it represent valuable tools for the fast and informative analysis of yeast microarray data.
DNA微阵列是在全基因组范围内获取表达数据的强大工具。我们进行了微阵列实验，以阐明转录网络，这些网络在酵母中因毒性多聚谷氨酰胺蛋白的表达而上调或下调。此类实验最初会生成包含差异表达基因的命中列表。为了研究转录反应，我们从这些基因构建了网络。因此，我们开发了一种算法，该算法能够通过基于从SPELL数据库获得的共调控关系对命中基因进行聚类来处理非常少量的微阵列。在这里，我们根据几个标准评估该算法，并进一步开发其统计能力。首先，我们定义了来自SPELL的共调控基因数量和输入命中基因数量如何影响网络质量。然后，我们展示了我们的网络准确预测更多差异表达基因的能力。将这些预测基因纳入网络可提高网络质量，并允许基于新实施的评分方法量化网络的预测强度。我们发现这种方法对我们自己的实验数据集以及我们从SPELL微阵列数据库测试的许多其他数据集都很有用。此外，通过所述算法获得的聚类极大地改善了对各个聚类的生物学过程和转录因子的分配。因此，将通过ClusterEx网络界面提供的所述聚类方法及其衍生的评估参数是对酵母微阵列数据进行快速且信息丰富的分析的宝贵工具。
