Structure-based prediction of ligand-protein interactions on a genome-wide scale

Abstract

We report a template-based method, LT-scanner, which scans the human proteome using protein structural alignment to identify proteins that are likely to bind ligands that are present in experimentally determined complexes. A scoring function that rapidly accounts for binding site similarities between the template and the proteins being scanned is a crucial feature of the method. The overall approach is first tested based on its ability to predict the residues on the surface of a protein that are likely to bind small-molecule ligands. The algorithm that we present, LBias, is shown to compare very favorably to existing algorithms for binding site residue prediction. LT-scanner's performance is evaluated based on its ability to identify known targets of Food and Drug Administration (FDA)-approved drugs and it too proves to be highly effective. The specificity of the scoring function that we use is demonstrated by the ability of LT-scanner to identify the known targets of FDA-approved kinase inhibitors based on templates involving other kinases. Combining sequence with structural information further improves LT-scanner performance. The approach we describe is extendable to the more general problem of identifying binding partners of known ligands even if they do not appear in a structurally determined complex, although this will require the integration of methods that combine protein structure and chemical compound databases.

Keywords: drug off-targets; machine learning; protein–ligand interactions; structure-based prediction.

Fig. 1.

Overview of LBias and LT-scanner methods. (A) For a given query protein (shown in green) LBias collects and superposes structure neighbors A, B, and C (shown in yellow) that are cocrystalized with their bound ligands (shown in blue). Then LBias predicts the most likely ligand-binding residues (shown in red and yellow) on the query protein based on collective contact information that the superposed ligands make. (B) For a given template cocrystal structure of a drug (shown in blue) and a template protein (shown in green), LT-scanner scans through protein A, B, and C (shown in orange) for by superposing the template structure onto each protein so as to create interaction models. Then LT-scanner calculates the Sim_LT-scanner interaction similarity score (shown as Sim_LT) between the interaction models of the query–drug complex and the interactions in the binding site of the template.

Fig. 2.

Precision–recall curves for ligand-binding residue prediction. Precision–recall curves are shown for LBias (black line), “simple count” (gray line), ConCavity (blue line), and random prediction (yellow line) in precision–recall curve space. Precision–recall points (PR-point; Results) are shown for LBias, ConCavity, simple count, COACH, and FTsite.

Fig. 3.

ROC curves for drug target protein predictions. (A) ROC curves for LT-scanner (black line), LT-scanner/seq(red line), and Sequence (green line) were shown to evaluate performance for prediction of drug targets in the full set of ∼15,000 human proteins in HSSP. (B) ROC curves for LT-scanner/seq (red line), FINDSITE_comb(30) (blue line), and FINDSITE_comb(95) (gray line) for drugs and proteins used in both the FINDSITE study and this study. (C) ROC curves for LT-scanner/seq (red line), LT-scanner (black line), Sequence (green line), and random (purple line). Curves calculated for 21 FDA-approved kinase inhibitors and 600 human kinases.

Fig. 4.

Calculating LBias SIM score. A query protein Q is shown at Left with three atoms, q₁, q₂, and q₃, identified specifically. The second panel shows a ligand-containing protein NL, structurally similar to Q with the ligand shown as a gray line connecting two gray atoms. Ligand-binding residues, n₁ and n₂, are identified. The NL complex is superposed onto Q. Residues from Q that interact with the superposed ligand are identified (q₁ and q₂ in the above). A protein–ligand similarity score between QL and NL (S_QL:NL) is then calculated, where $S_{QL:NL}=m_{11}{e}^{-\gamma r_{11}^2}+m_{12}{e}^{-\gamma r_{12}^2}+m_{21}{e}^{-\gamma r_{21}^2}+m_{22}{e}^{-\gamma r_{22}^2}$. S_QL:NL is a function of all of the pairwise distances (e.g., r₁₁, r₁₂) between the atoms from Q interacting with the superposed ligand and the atoms from N interacting with the native ligand if the two atoms in question make chemically similar contacts with the ligands. For example, if n₁ makes a hydrogen bond with the ligand, but q₁ is hydrophobic, m₁₁ would be zero and there would be no contribution to S_QL:NL from this pair of atoms (Materials and Methods).