Protein expression profiling

Protein expression profiling is defined in general as identifying the proteins expressed in a particular tissue, under a specified set of conditions and at a particular time, usually compared to expression in reference samples. This information is useful in drug discovery and diagnosis as well as in understanding response mechanisms at the protein level. We may identify all the proteins responding to a particular stimulus and select those whose expression changes most. Or we may isolate significant protein variables and then identify them. These definitive sets of proteins (protein expression signatures; PES) are specific to diseases, toxicants, physical stresses, and to degrees of stress severity. Here we describe a method, based on machine learning, for isolating the sets of proteins, before identifying them by name, which classify accurately the treatment classes in a study. The principle in this chapter is that if proteins associated with known classes of interest can be used to identify unknown classes then the proteins are definitive for diagnosis.The proteins in each class, including controls, are converted to digital data and serve as input to artificial neural network (ANN) models. Multiple two-dimensional electrophoresis (2DE) gel patterns are included in each treatment class. A training subset of digitized individual, not composite, gel images is used to construct an ANN model which is then applied to a test set of images. Successful classification of the unknown (test) data confirms that the variables included in the model are indeed significant in discrimination among the classes. In the study described here the misclassifications were 5% or less using the ANN models. The ANN method seems to be a useful complement to image analysis, described in Chapter "Troubleshooting Image Analysis in 2DE". The reduction in protein variables permits multivariable statistics such as cluster and discriminant analyses.