基因芯片数据标准化局部加权回归法权函数探究

局部加权回归方法(Lowess方法)作为一种较常见的基因芯片数据标准化方法涉及到众多的参数和权函数.对于权函数的选择一直以来都是采用三次权函数.因此本文从误差为重尾型的t分布出发,模拟不同自由度的基因芯片数据,然后选择不同的权函数对这些数据进行Lowess方法标准化,并利用残差平方和、标准化后的M和A值的相关系数及MA图作为评判标准,最终得到相关指标量与权函数的变化关系.

Normalization for microarray data: the weight function of local weight regression

Local weight regression (Lowess) is a kind of widely used method, which need used to be given a lot of parameters and weight function. When this method is used to normalize data, thrice weight function is always used. So we want to know about the situation in which the noise of data obeys a distribution with heavy tail occurs what kind of weight function should we select. First the simulation data and the noise obeys t distribution with different degree of freedom are given, then the data by the Lowess method with different weight function is normalized, and finally sum of residual, the correlation between M after normalization and A, MA-plot are used to evaluate the weight function, and their relationship is gained.