Structural Bioinformatics Group
Division of Molecular Biosciences
Faculty of Natural Sciences


3DLigandSite -Ligand binding site prediction Server

Home     About     News     Help     FAQ     Example     Contact     Disclaimer
3DLigandSite Overview

Summary

3DLigandSite uses predicted protein structures and the ligands present in homologous structures to predict ligand binding sites. An overview of the method used by 3DLigandSite is shown on the left. The server and method are described in the NAR 2010 Web server edition[1]. 3DLigandSite is based on our human method that wsa one of the best predictors of binding sites at CASP8[2][3].

 

Detailed Method

1. Structure Prediction

Users can either submit a query sequence or a structure. When a sequence is submitted, the first step is to predict the protein struture using the Phyre server[4]. This step is not necessary if the user has provided their own structure.

2. Searh Structural Library of Ligand bound structures

The next step is to identify structures homologous to the query that have ligands bound. MAMMOTH [5] is used to perform a full structural scan of the modelled structure against a library of protein structures with bound ligands. Upto the top 25 scoring (using MAMMOTH -lnE score) are retained for analysis. These structures are aligned with the modelled structure using TMalign [6]. If we remove the homologus structures but retain their ligands, the ligands are superimposed onto our modelled structure.

3. Ligand Clustering & binding site prediction

Single linkage clustering is used to group the ligands. The cluster with the most ligands is selected and residues that are within a distance threshold of the clustered ligands are predicted to form part of the binding site. Residue conservation is calculated unsing the Jensen Shannon divergence score [7] and provided as a guide for the user. Residue conservation is not used as part of the 3DLigandSite predictive process. Conservation information can be used in conjunction witht the 3DLigandSite prediction.

Rerences

[1] Wass MN, Kelley LA & Sternberg MJ (2010) 3DLigandSite: predicting ligand-binding sites using similar structures. NAR 38 Suppl:W469-73. PubMed.
[2] Wass, M.N. and Sternberg, M.J. (2009) Prediction of ligand binding sites using homologous structures and conservation at CASP8. Proteins, 77 Suppl 9:147-51.
[3] López G., Ezkurdia I., Tress M.L. (2009) Assessment of ligand binding residue predictions in CASP8. Proteins, 77 Suppl 9:138-46.
[4] Kelley L.A. and Sternberg M.J. (2009) Protein structure prediction on the web: a case study using the Phyre server. Nature Protocols 4:363-71.
[5] Ortiz, A. R., C. E. Strauss, and O. Olmea (2002). Mammoth (matching molecular models obtained from theory): An automated method for model comparison. Protein Sci 11:2606-21.
[6] Zhang, Y. and Skolnick, J. (2005) TM-align: A protein structure alignment algorithm based on TM-score , Nucleic Acids Research, 33: 2302-09.
[7] Capra, J. & Singh, M. (2008) Characterization and prediction of residues determining protein functional. Bioinformatics, 24, 1473-1480.

Submission Options

The 3DLigandSite submission page has been kept as simple as possible, users can either submit a query sequence here or a protein structure here. Email address is options, users providing an email address will receive an email confirming submission a further message once the job is complete. Job description is also optional and this column allows the use to assign their own description to the job. This is particularly useful if a user is submitting multiple jobs.

The 3DLigandResults Page

Submission Details
Email: m.wass04@imperial.ac.uk
Unique Job identifier: example4
Description: T0406
Date: Wed Sep 21 09:29:35 GMT 2009
Submission Type: sequence
Query Seq: GLAACEGEYSQKYSTMSPLGSGAFGFVWTAVDKEKNKEVVVKFIKKEKVLEDCWIEDPKL
GKVTLEIAILSRVEHANIIKVLDIFENQGFFQLVMEKHSGLDLFAFIDRHPRLDEPLASY
IFRQLVSAVGYLRLKDIIHRDIKDENIVIAEDFTIKLIDFGSAAYLERGKLFYTFCGTIE
YCAPEVLMGNPYRGPELEMWSLGVTLYTLVFEENPFCELEETVEAAIHPPYLVSKELMSL
VSGLLQPVPERRTTLEKLVTDPWVTQPVNLADYTWEEVFRVNKPESGVLSAASLEMGNRS
LSDVAQAQELCGGE

 

While a job is running, the results page displays the status of the job indicating where in the predictive process the job currently is. The submission details are also displayed an example is shown on the right. This includes all of the basic information associated with the submission including the unique job identifer which is required to retrieve results from the server.

Click here to view the results page of this job.

Structural Model
Phyre job: 8ce9f8caffc285eb
Phyre2 template:3dlsA_
Phyre2 confidence score: 100.0
Structural Search
confidence data from search of structural library with Mammoth
Average lnE:29.965
Maximum LnE:34.64
Min LnE:28.45

Details of the structural model used for the prediction are also displated. When a sequence has been submitted the Phyre job id is disaplayed (which is normally the same as the 3DLigandSite job id) and this link can be used to see the Phyre job progress and results. The top Phyre template is displayed, where the second to fifth characters indicate the pdb structure that the template is from and the 6th character the chain in the structure. The e-value of the phyre prediction is also displayed, to give an indication to the user of how confident the structure prediction is. Warning messages are displayed when a low confidence model is used so that the user is aware and can consider this when using the 3DLigandSite prediction.

Some overview details of the structural search against the library of ligand bound structures is used to indicate how close the structural hits are to the modelled query structure. Of the the hits that have been selected for use their average, maximum and minimum -lnE scores are displayed. A -lnE score of 4 suggests that the two proteins are likely to be similar, the higher the score the closer the match. Only structures with a -lnE score of 7 or greater a included in the analysis.

 

clusterTable

The Cluster Table

This table displays details of the clusters of ligands identified on the target model. In this example only a single cluster was identified. When there are multiple clusters they can be clicked on to display a page showing details of the residues they contact and a Jmol applet simialr to the one on the resutls page.
3DLigandSite Overview

Predicted Binding Site Table

The residues predicted to form part of the binding site are shown in the left table. For each residue its residue number and amino acid code is displayed. The other columns display numbers relating to the ligands that are in contact or close to the residue as described below:

contact - The number of ligands that the residues is within 0.8 anstrom separation (including their Van der Waals radii). e.g. Residue 48 contacts 9 ligands.

av distance - The average separation of the residue from all the ligand that are within 0.8 angstroms. e.g. for residue 48 the 9 ligands are have an average separation from the residue of 0 angstroms (i.e. they are touching).

JS divergence - the conservation score for the residue calculated usign Jensen Shannon divergence. maximum score is 1 and minimum score is 0. So the higher the score the more conserved the residue is. The conservation is not used in the predictive process but it may be helpul to the user when interpreting the prediction.

Heterogen Table

The heterogen table lists the heterogens that are present in the ligand cluster that is being used for the prediction. The number of each type of ligand and the structures they originated from are also displayed

 

Using Jmol to view the prediction

see description below image - view these results here

3DLigandSite Overview

The left window is a Jmol applet which displays the modelled structure. The rediues predicted to form part of the binding site are coloured blue. The lilgands that form the cluster used for the prediction are also displayed with metla ions shown in spacefill format and non metal ligands as wireframes. The JAVA runtime environment is required to use the Jmol applet this and is freely avalable from www.java.com if it is not already installed on your system. Documentation on using JMOL can be found at www.jmol.org.

The right panel provides the user with a range of viewing options to modify the display of the protein, binding site prediction and ligands in the Jmol applet. The top section controls the display of the whole protein, allowing either a cartoon, spacefill or wireframe representation. The residues can be coloured according to the prediction or by their conservtion score (Jensen Shannon divergence) a key at the bottom of the panel provides a guide for the residue colouring.

Further options are availble to modify the display of the residues predicted to form the binding site and the ligands used to make the prediction. Further the protein can be coloured to show the predicted binding site or the Jensen Shannon divergence score. Predicted residues can also be labelled for easier identification. Finally the user can change the background colour of the Jmol applet and set the protein to rotate. Click to view examples of the way the display can be modified.

 

 

Home     About     News     Help     FAQ     Example     Contact     Disclaimer
© Structural Bioinformatics Group, Imperial College, London
Mark Wass