Protein Structure and Function: Application of Bioinformatics Methods - John Rigden 2014

Prediction of Membrane Protein Structure
Prediction of Transmembrane Protein Topology
Beta-Barrel Proteins

The number of alpha-helical TM Proteins, both in complete proteomes and in 3D Structure Databases, is relatively large compared to the number of proteins with a beta-barrel structure. Consequently, Methods for predicting the structure and topology of Beta-barrels are developing less intensively. Another reason for this disparity in methodological development is likely the relative ease of predicting The structure of TM alpha-helices due to the high proportion of hydrophobic residues they contain. In contrast, the antiparallel beta-strands of TM beta-barrels feature alternating polar and hydrophobic Amino Acids. This architecture directs hydrophobic residues toward the membrane, while polar residues face the solvent-exposed surface. Early methods for predicting beta-strand topology relied on sliding-window Hydrophobicity analysis to identify alternating structural elements (Schirmer and Cowan 1993). Other approaches employed specialized empirical rules based on amino acid propensities and the structural characteristics of proteins (Gromiha and Ponnuswamy 1993). As the number of beta-barrels with atomic-resolution structures increased, machine learning-based methods began to emerge. These include neural networks (Jacoboni et al. 2001; Gromiha et al. 2004), Hidden Markov Models (Martelli et al. 2002; Liu et al. 2003; Bagos et al. 2004), and support vector machine-based prediction (Park et al. 2005), which utilize single and Multiple Sequence Alignments. Table 4.4 lists A number of machine learning-based methods for predicting beta-barrel structure and topology.

Class="center">Table 4.4. Machine learning-based methods for predicting transmembrane beta-barrel topology

* - topology prediction is performed using multiple sequence alignments (MSA).

** - the method is suitable for whole-Genome Analysis (WGA)