LigSearch: a knowledge-based web server to identify likely ligands for a protein target

Abstract

Identifying which ligands might bind to a protein before crystallization trials could provide a significant saving in time and resources. LigSearch, a web server aimed at predicting ligands that might bind to and stabilize a given protein, has been developed. Using a protein sequence and/or structure, the system searches against a variety of databases, combining available knowledge, and provides a clustered and ranked output of possible ligands. LigSearch can be accessed at http://www.ebi.ac.uk/thornton-srv/databases/LigSearch.

Keywords: LigSearch; ligand prediction.

Figure 1

A schematic outline of the workflow in LigSearch. The user can submit either a structure or a sequence.

Figure 2

Validation results for LigSearch runs on 200 randomly selected enzymes with no three-dimensional structural model in the PDB. (a) Histogram of the molecular-similarity scores for the closest PDB ligand match, as computed by the SMSD program, to any of the enzyme’s cognate ligands. (b) Histogram of the sequence identities between the query enzyme sequence and the PDB protein from which the best ligand match has a similarity of 0.8 or greater to one of the cognate ligands. (c) Histogram of the LigSearch scores for the best matches to cognate ligands. The counts are grouped into four sets according to the similarity score, s, of the best-matching molecule. Lowest similarity scores (s < 0.7) are shown in blue, scores 0.7 ≤ s < 0.8 are shown in green, scores 0.8 ≤ s < 0.9 are shown in orange and closest matches with s ≥ 0.9 are shown in red. (d) Histogram of counts of molecules with similarity s ≥ 0.8 to at least one of the enzyme’s cognate ligands as returned by LigSearch for the non-PDB hits. The cognate molecules themselves are, of course, excluded from the results.

Figure 3

A plot of the top-scoring molecules in each of the 47 clusters returned by LigSearch for UniProt entry P65248. The molecules have been laid out using multi-dimensional scaling on the basis of their all-by-all similarities. Thus, similar molecules tend to be grouped together. The labels show the cluster number in square brackets and the PDB Het Group three-character name or ChEBI identifier. Red labels correspond to molecules from matches to PDB entries, while blue labels are molecules returned by ChEBI searches. The molecular diagrams were plotted using ChemDraw (http://www.cambridgesoft.com).

Figure 4

Three cluster representatives for the molecules listed in Table 1 ▶. The molecules are annotated according to the interactions that they make with the protein in the top-scoring PDB entry for the cluster. Atoms making hydrogen bonds to protein are depicted with spokes radiating from them, while hydrophobic interactions have a grey circle around them (none in this example). The colour of the spokes corresponds to the similarity of the residue to which the hydrogen bond is made and the corresponding residue in the query protein (which in this case is thymidylate synthase from S. aureus; UniProt ID P65248): red for identical residue type, orange for similar and dark grey for different. The images are provided in the results section for every query with PDB hits. The molecules and the PDB entries from which the data come are (a) P ¹-(5′-adenosyl)-P ⁵-(5′-thymidyl)pentaphosphate (PDB entry 4tmk), (b) thymidine 5′-diphosphate (PDB entry 3hjn) and (c) 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (PDB entry 2pir).

Figure 5

A schematic diagram of the protein–ligand interactions in two distantly related proteins: (a) thymidylate synthase from S. aureus (PDB entry 4eaq) and (b) human thymidylate synthase (PDB entry 1e99). The ligands (blue bonds) in both are identical: 3′-azido-3′-deoxythymidine 5′-­monophosphate. Equivalent protein residues in the two plots are circled in red and occupy the same positions in each plot: for example, Glu37 is equivalent to Phe42, Phe66 is equivalent to Phe72 and Tyr100 is equivalent to Phe105. Hydrogen bonds are depicted by green dotted lines and labelled with their length in Å, while hydrophobic interactions are represented by red arcs whose spokes radiate towards the ligand. The diagram was generated using LigPlot ⁺ (Laskowski & Swindells, 2011 ▶).