Transmembrane β-barrels are embedded in the outer membrane of Gram-negative bacteria, mitochondria and chloroplasts. The cellular location and functional diversity of β-barrels makes them an important protein class.

At the present time, very few β-barrel structures have been determined by X-ray diffraction because of the experimental difficulty encountered in crystallizing transmembrane proteins. Despite current paucity of data, the only prediction methods currently available, with the exception of the method of this paper, are based on machine learning methods. Such machine learning methods cannot use critical information from inter-residue contacts, even though it is known that such contacts are essential in folding.

Here, we introduce a novel method to predict the supersecondary structure of β-barrel proteins, and use this method to perform an in silico study of folding properties of transmembrane β-barrels. Rather than using machine learning (hidden Markov models, neural networks, support vector machines, etc.), our method is based on an underlying energy model, whose parameters come from an analysis of globular proteins (i.e. independent of outer membrane proteins). Our algorithm applies multi-tape S-attribute grammars to describe potential β-barrel supersecondary structure and then computes by dynamic programming the minimum free energy β-barrel structure. Hence, our approach effectively uses long-range interactions to compute the optimal transmembrane β-barrel structure.

transFold is more than a simple prediction tool. By modification of the parameters defined in the physical approximate model, we are able to study of the folding properties of a polypeptide. This leads transFold to be an effective tool for theoretical biology.