Protein annotation and modelling servers at University College London

Abstract

The UCL Bioinformatics Group web portal offers several high quality protein structure prediction and function annotation algorithms including PSIPRED, pGenTHREADER, pDomTHREADER, MEMSAT, MetSite, DISOPRED2, DomPred and FFPred for the prediction of secondary structure, protein fold, protein structural domain, transmembrane helix topology, metal binding sites, regions of protein disorder, protein domain boundaries and protein function, respectively. We also now offer a fully automated 3D modelling pipeline: BioSerf, which performed well in CASP8 and uses a fragment-assembly approach which placed it in the top five servers in the de novo modelling category. The servers are available via the group web site at http://bioinf.cs.ucl.ac.uk/.

Figure 1.

The diagram produced by the MEMSAT-SVM algorithm available via the PSIPRED server. (a) A schematic diagram of the transmembrane regions is presented. Directly below this a trace of the Kyte–Doolittle hydropathy plot (18) and below that four traces of the outputs for the four SVMs used to make the transmembrane region assignment are shown. (b) A cartoon of the transmembrane helix topology summarizing the linear coordinates for the helices and indicating where the protein’s extra- and intercellular regions are.

Figure 2.

The process flow when analysing a sequence using the BioSerf pipeline is shown. An incoming sequence is initially analysed by both PSI-BLAST, PSIPRED and pGenTHREADER. The PSI-BLAST run allows the pipeline to identify whether the query sequences closely matches the sequence of a known structure in the PDB. PSIPRED and pGenTHREADER attempt to identify any remote structural homologues of the sequence should the PSI-BLAST run fail to find any related sequences. These data are then combined and potential template structures for modelling are selected given suitable cut-offs. Where there is at least one suitable template, the pipeline then uses MODELLER to generate a homology model. Where there are no suitable template structures, FRAGFOLD is used to build a possible model structure. A typical run of the pipeline that finds a suitable template structure takes around 10 min. In those cases where no template can be found and FRAGFOLD is used, runs can take up to 3 h.

Figure 3.

A typical series of analyses using our servers to annotate a protein of unknown function are shown. Putative human membrane protein A8MYE5 was selected for this analysis (a). It was initially submitted to a MEMSAT-SVM prediction to locate any transmembrane helices (b). In the third step, (c), the sequence is submitted to DomPred to locate any possible domain boundaries. As these do not conflict with the transmembrane helix assignment, the sequence is divided into it’s possible constituent domains and those domains that are not transmembrane are submitted to BioSerf for homology modelling (d). The larger domain has a long terminal region consisting of two beta sheets that are likely to be disordered rather than packed as in the prediction. To confirm this possibility, the larger domain is submitted the DISOPRED2 server. The DISOPRED2 prediction (e) confirms that the terminal stretch of the larger domain is likely to be disordered rather than packed onto the structure.

Figure 4.

The final, putative, annotation of Uniprot A8MYE5 sequence.