Getting Docking into Shape Using Negative Image-Based Rescoring

Abstract

The failure of default scoring functions to ensure virtual screening enrichment is a persistent problem for the molecular docking algorithms used in structure-based drug discovery. To remedy this problem, elaborate rescoring and postprocessing schemes have been developed with a varying degree of success, specificity, and cost. The negative image-based rescoring (R-NiB) has been shown to improve the flexible docking performance markedly with a variety of drug targets. The yield improvement is achieved by comparing the alternative docking poses against the negative image of the target protein's ligand-binding cavity. In other words, the shape and electrostatics of the binding pocket is directly used in the similarity comparison to rank the explicit docking poses. Here, the PANTHER/ShaEP-based R-NiB methodology is tested with six popular docking softwares, including GLIDE, PLANTS, GOLD, DOCK, AUTODOCK, and AUTODOCK VINA, using five validated benchmark sets. Overall, the results indicate that R-NiB outperforms the default docking scoring consistently and inexpensively, demonstrating that the methodology is ready for wide-scale virtual screening usage.

Figure 1

Negative image-based rescoring step-by-step. The negative image-based rescoring (R-NiB) protocol follows five steps: (1) Ligand-binding cavity and its centroid are selected from the protein 3D structure (a cartoon model of RXRα with the bound docosa hexaenoate or HXA; PDB: 1MV9). (2) Negative image or NIB (negative image-based) model (transparent surface), composed of neutral filler atoms (cyan spheres) and negative cavity point (red sphere), is generated using PANTHER. (3) Flexible molecular docking (e.g., VINA) is performed for the ligands (e.g., lig. #1 and lig. #2 or C44184559 and CHEMBL2085503 in the DUD-E set for the RXRα), and several (e.g., N = 3) alternative docking poses (stick models with white backbone) are outputted for each compound. (4) Cavity-based rescoring or the shape/charge comparison of docking poses (one at a time!) is used with the NIB model without geometry optimization using ShaEP. (5) Comparison produces similarity scores (from 1 to 0) for each docked pose, and this information is used to rank the individual docking poses and the ligands. Based on the R-NiB ranking, compounds can be categorized or predicted as inactive (red stick model; e.g., lig. #1) or active (green stick model; e.g., lig #2). Note that the steps involved in the protein or ligand preparation for NIB model generation or docking, visual inspection of the best-ranked poses, or potential benchmarking efforts are omitted for brevity. The figure was created using BODIL and Visual Molecular Dynamics or VMD 1.9.2.

Figure 2

Negative images of target proteins’ ligand-binding cavities. On the left, 3D structures of target proteins (cartoon models) with co-crystallized ligands (CPK models) at binding cavities. In the middle, cross sections of binding cavities (opaque surfaces) in close ups (red boxes). PANTHER-generated negative images or NIB (negative image-based) models are composed of neutral filler atoms and negatively or positively charged cavity points (cyan/red/blue spheres). On the right, NIB models are shown with space-filling transparent surfaces either with cavity points or an active ligand from PLANTS docking (sticks with a green backbone). Both the shape and volume (N = 44–79) of NIB models vary substantially between (A) mineralocorticoid receptor (MR; PDB: 2AA2), (B) neuraminidase (NEU; PDB: 1B9V), (C) retinoid X receptor alpha (RXRα; PDB: 1MV9; different angle shown in Figure 1), (D) cyclooxygenase-2 (COX-2; PDB: 3LN1), and (E) phosphodiesterase-5 (PDE5; PDB: 1UDT). NIB models aim to encompass only those cavity sections needed for ligand binding instead of filling the cavities to the brim. The figure was created using BODIL and Visual Molecular Dynamics or VMD 1.9.2.

Figure 3

Docking and rescoring performance as receiver operator characteristic curves. The linear receiver operator characteristic (ROC) curves are plotted for the original docking and rescoring results of negative image-based rescoring (R-NiB; Figure 1) or SMINA rescoring. Benchmarking is shown for a selected assortment of results, but the full set of data is given in the Supporting Information for each software and test set in the form of linear (Figures S1–S5) and semilogarithmic ROC curves (Figures S6–S10). The figure was created using ROCKER0.1.4.

Figure 4

Negative image-based rescoring refocuses neuraminidase docking. (A) Best poses of 10 top-ranked docked compounds (stick models), sampled and selected by docking algorithm PLANTS, are relatively scattered in the partially open surface pocket of neuraminidase (NEU; opaque magenta surface). The cavity center (green sphere), which is the geometric centroid of the co-crystallized ligand BANA206 (PDB: 1B9V), was used to center PLANTS docking and NIB model generation with PANTHER. (B) The NIB model (transparent yellow surface), which was used in the cavity-based rescoring with ShaEP, is shown with the centroid. (C) Best poses of 10 top-ranked docked compounds selected by negative image-based rescoring (R-NiB; Figure 1) form a much tighter cluster than scattered ligands/poses selected at the top by default docking scoring of PLANTS. The figure was created using BODIL and Visual Molecular Dynamics or VMD 1.9.2. See Figure 2 for more information.