Ferhatosmanoglu Nilgun, Allen Theodore T, Catalyurek Umit V
Int J Data Min Bioinform. 2015;13(1):31-49. doi: 10.1504/ijdmb.2015.070838.
Two-colour microarrays are used to study differential gene expression on a large scale. Experimental planning can help reduce the chances of wrong inferences about whether genes are differentially expressed. Previous research on this problem has focused on minimising estimation errors (according to variance-based criteria such as A-optimality) on the basis of optimistic assumptions about the system studied. In this paper, we propose a novel planning criterion to evaluate existing plans for microarray experiments. The proposed criterion is 'Generalised-A Optimality' that is based on realistic assumptions that include bias errors. Using Generalised-A Optimality, the reference-design approach is likely to yield greater estimation accuracy in specific situations in which loop designs had previously seemed superior. However, hybrid designs are likely to offer higher estimation accuracy than reference, loop and interwoven designs having the same number of samples and slides. These findings are supported by data from both simulated and real microarray experiments.
双色微阵列用于大规模研究差异基因表达。实验规划有助于减少关于基因是否差异表达的错误推断的可能性。此前针对此问题的研究主要集中在基于对所研究系统的乐观假设，依据诸如A最优性等基于方差的标准来最小化估计误差。在本文中，我们提出了一种新颖的规划标准来评估微阵列实验的现有方案。所提出的标准是“广义A最优性”，它基于包括偏差误差在内的现实假设。使用广义A最优性，在特定情况下，参考设计方法可能会产生更高的估计精度，而在这些情况下，循环设计此前似乎更具优势。然而，混合设计可能比具有相同样本数量和载玻片数量的参考、循环和交织设计提供更高的估计精度。这些发现得到了来自模拟和实际微阵列实验数据的支持。
