Abstract

The probability of expression (i.e., POE) scale was developed to achieve two main goals: (1) Microarray data are generated using a variety of measurement techniques that are not directly comparable. We sought to develop a common scale for which microarray data generated using differing methods could be converted and then compared. (2) In many cases, we are interested in defining whether genes and/or samples fall into one of three categories: overexpressed, underexpressed, or normally expressed. However, gene expression values are usually generated on a continuous scale. The scale that we have developed is categorical, assigning these continuous individual expression values probabilities of falling into one of these three categories. We describe the POE scale, several of its practical uses, and demonstrate its use on a lung cancer microarray data set.

POE: A Latent Variable Mixture Model

The Motivation and Practicality of POE

One reason for developing the probability of expression (POE) scale is to transform continuous expression values to a three-component categorical scale. Often the continuous expression values are displayed by image plots that use a red–green scale for over- and underexpression, respectively. The commonly used red and green image plots effectively display microarray data and generally have three main colors: red (overexpression), green (underexpression), and black (normal expression). The visual impact is that we see red, green, and black and not the “smooth-scale” that would blend from red to black to green as implied by continuous data. Our goal is to assign probabilities to each observed expression value where the probabilities correspond to the chance that an expression value should be called “red,” “green,” or “black.”