Prediction of membrane proteins based on classification of transmembrane segments.

Abstract

The number of transmembrane segments often corresponds to a structural or functional class of membrane proteins such as to seven-transmembrane receptors and six-transmembrane ion channels. We have developed a new prediction method to detect the membrane protein class that is defined by the number of transmembrane segments, as well as to locate the transmembrane segments in the amino acid sequence. Each membrane protein class is represented by a model of ordering different types of transmembrane segments. Specifically, we have classified the transmembrane segments in known membrane proteins into five groups (types) using the Mahalanobis distance with the average hydrophobicity and the periodicity of hydrophobicity as a measure of similarity. The discriminant functions derived for these groups were then used to detect transmembrane segments and to match with the models for one- to fourteen-spanning membrane proteins and for globular proteins. Using the test data set of 89 membrane proteins whose transmembrane positions are known by experimental evidence, 61.8% of the proteins and 85.1% of the transmembrane segments were correctly predicted. Because of the new feature to predict membrane protein classes, the method should be useful in the functional assignment of genomic sequences.