ESTIMATION OF SIGNAL THRESHOLDS FOR MICROARRAY DATA USING MIXTURE MODELING

摘要

The application relates to a method of classifying DNA microarray intensity data into two classes of reliable and unreliable data, comprising the steps of extracting median and mean intensities of two fluorescent signals of different wavelengths from cDNA sequences arranged in a microarray as lambda1-channel and lambda2lambda-channel raw data; estimating normal mixture model parameters corresponding to class conditional densities and prior probabilities using expectation maximization algorithm; determining in univariate analysis an optimal signal threshold for each channel or in bivariate analysis a threshold surface for said microarray raw data using Bayesian classification theory; denoting a signal intensity value as reliable, if either of the channels is larger than its respective threshold in univariate analysis or if the 2-dimentional data point falls outside of a hyper-quadratic decision region in bivariate analysis.