History

J. Waldispühl developed the general approach of multi-tape S-attribute grammars for the structure prediction of both TM α-helical bundles and of TM β-barrels, as part of his Ph.D. dissertation in Computer Science at École Polytechnique under the supervision of J.-M. Steayert.

The energy model, grammar and dynamic programming algorithm for structure prediction of TM α-helical bundles appeared in (Waldispühl & Steyaert, 2005). Although the approach for TM β-barrels was described in the dissertation of J. Waldispühl, critical features of the energy model were subsequently implemented, resulting in state-of-the-art performance of the current transFold program, written and tested by J. Waldispühl while working as a postdoctoral associate with P. Clote and B. Berger. The contact potentials for transFold come from B. Berger.

Web Server

Although the transFold webserver uses the most recent implementation of the software described in the original papers cited in the references, the current interface does not provide all the capacities of the author's Linux distribution (there is some limitations in parameter settings and constrained secondary structure).

The web server allows one to stipulate the pore type as being either non water-filled or water-filled (see original paper for more details). Other parameters are automatically assigned default values by the web server; however the algorithm described in

• J. Waldispühl, B. Berger, P. Clote and J.-M. Steyaert,
  Predicting Transmembrane β-barrel Proteins and Inter-Strand Residue Contacts From Sequence
  To appear in PROTEINS: Structure, Function and Bioinformatics

provides one to input non-default parameters. The advanced submission form of the web server allows the user to input such parameters.

Time used in the computation made by the web server may vary, especially depending on the length of the protein. However, computations are usually fast residues, taking less than one minute. Depending on the protein sequence, transFold could require a large amount of memory. To avoid extensive uses of our computer cluster, input size is limited to 350 residues. Users who wish to run transFold on larger sequences are encouraged to send a request to Jérôme Waldispühl, or to purchase the Linux distribution license.