Pattern discovery in expression profiling data.

In expression profiling studies, it is often necessary to identify groups of genes with similar expression profiles in a variety of samples, and/or groups of samples with similar expression profiles. Each profile can be expressed as a single data point in a space with the same number of dimensions as there are parameters in the profiles. In this way, pattern discovery among expression profiles is translated into pattern discovery in the spatial distribution of data points: the similarity between profiles is defined by the distance between the corresponding data points. Various multivariate analysis methods, such as clustering and dimensionality reduction methods, are used to summarize the data point distribution to help the investigator recognize major trends. As different methods may identify different features of the distribution, it is important to analyze a particular data set with multiple methods.