Biometric tools for microarrays

来自微阵列的基因表达数据为理解一般的基因作用，特别是杂种优势开辟了广泛的选择。通常，在微阵列上发现1.000个基因并筛选差异表达，而重复的数量往往由于预算限制而受到限制，因此变量的数量远远超过重复的数量，过度拟合的危险甚至比传统的QTL定位更大。此外，表达数据受到无数随机变异来源的影响，从而模糊了潜在的表达模式。鉴于这些困难，考虑到所有变化来源的现实建模是至关重要的。该项目将为这项任务制定一个通用的混合建模和模型选择框架。分析早期阶段的一个关键问题是数据转换的选择。虽然对数变换因其对褶皱变化的简单解释而经常被使用，但也有人提出了其他的建议。这个项目将探索不同转换的特性和适用性。此外，还将包括乘法模型和因子分析模型，以解释不同基因之间的规模差异。将特别强调说明误差变异的所有相关来源(染料、载玻片、行和列、复制、现场和实验室试验变异等)。最后，解释和预测杂种优势的模型将被整合到一般的建模框架中。

Gene-expression data from microarrays open up a wide range of options for understanding gene action in general and heterosis in particular. Typically, many 1.000 genes are spotted on a microarray and screened for differential expression, while the number of replicates is often limited due to budget constraints, so the number of variables by far exceeds that of replications and the danger of over-fitting is even larger than in conventional QTL mapping. Also, expression data are influenced by a myriad of sources of random variation, thus obscuring the underlying pattern of expression. In light of these difficulties, realistic modelling that takes into account all sources of variation is of paramount importance. This project will develop a general mixed modelling and model selection framework for this task. A crucial question in the early stages of analysis is the choice of data transformation. While the log-transformation has been frequently used because of its simple interpretation with respect to fold change, others have been suggested. This project will explore the properties and suitability of different transformations. In addition multiplicative and factor-analytic models will be included to account for scale differences among different genes. Particular emphasis will be given to accounting for all relevant sources of error variation (dyes, slides, rows an columns, replications, field and laboratory trial variation, etc.). Finally, models for explaining and predicting heterosis will be integrated into the general modelling framework.