Volume 26, Issue 3, Pages 499-512.e2 (March 2018) An Augmented Pocketome: Detection and Analysis of Small-Molecule Binding Pockets in Proteins of Known 3D Structure  Raghu Bhagavat, Santhosh Sankar, Narayanaswamy Srinivasan, Nagasuma Chandra  Structure  Volume 26, Issue 3, Pages 499-512.e2 (March 2018) DOI: 10.1016/j.str.2018.02.001 Copyright © 2018 Elsevier Ltd Terms and Conditions

Structure 2018 26, 499-512.e2DOI: (10.1016/j.str.2018.02.001) Copyright © 2018 Elsevier Ltd Terms and Conditions

Figure 1 Flowchart Workflow illustrating the analysis of (A) identification of ligand binding pockets from PDB. Three different algorithms were used for detecting pockets and only those that are detected by all three were considered for further analysis; (B) delineating whether a given pocket location is known or new; (C) Exhaustive comparison of sub-structures leading to ligand associations and structural annotation; and (D) specific application scenarios of the annotation. RCSB, Research Collaboratory for Structural Bioinformatics; SG, structural genomics. Structure 2018 26, 499-512.e2DOI: (10.1016/j.str.2018.02.001) Copyright © 2018 Elsevier Ltd Terms and Conditions

Figure 2 Representation of the Augmented Pocketome (A) A schematic illustrating consensus pocket detection for a given protein. It can be seen that only those pockets predicted by all three are considered in the pocketome. (B) Fold coverage of the pocketome showing that the 1,076 folds covered by the pocketome are about twice those covered by known pockets. (C) A schematic showing the new methodology adopted for labeling a pocket location as known or new, by considering all proteins in the family of a given protein P1. Protein P1 has been detected with six pocket locations, of which K1, K2, and K3 are known locations in one member or the other, and hence can be extrapolated from homologous proteins (P2, P3, P4, and P5), while N1–N3 are new pocket locations. (D) A bar chart showing the number of known versus number of new pockets, suggesting that the augmented pocketome is about seven times greater than what was thought. Structure 2018 26, 499-512.e2DOI: (10.1016/j.str.2018.02.001) Copyright © 2018 Elsevier Ltd Terms and Conditions

Figure 3 Overview of the Site-Comparison Analysis (A) Workflow describing the dataset preparation of known binding sites (KS) from PDB, KnownSitesDB, which consisted of 84,839 redundant sites. Site comparison of 249,096 pockets in the pocketome (P1–Pn) versus KnownSitesDB using PocketMatch leading to ligand annotation. (B) Pie chart (left) showing the distribution of known and new pockets containing ligand hits (blue sectors), of which 10% are already known in PDB, and 46% are the pockets that resemble the known ones. The rest 44% (in yellow) represent those pockets that do not show any similarity to known binding sites and hence currently have no ligand associations. The right panel shows the number of pocket similarities at different PocketMatch thresholds. (C) A stacked plot showing the range of residue matches at different PocketMatch thresholds. x axis shows different bins of the number of residues that are matched, y axis indicates the number of pockets showing similarity with known ones at different PocketMatch thresholds shown in different colors. (D) Example alignments for two new pockets (blue) showing significant similarities with known sites (red). Ligand originally bound to the structure in the known site (red) is shown in ball and stick model colored by atom type. Structure 2018 26, 499-512.e2DOI: (10.1016/j.str.2018.02.001) Copyright © 2018 Elsevier Ltd Terms and Conditions

Figure 4 Pocket Comparison (A) Number of site types obtained at different PocketMatch thresholds. Although a broad classification using PMSmax 0.5 resulted in 1,293 clusters, a finer level of classification, best optimized for clustering pockets, at PMSmax 0.7 yielded a total of 2,161 clusters, which amounts to the types of sites present in the pocketome. (B) A portion of the whole network obtained at PMSmax 0.7. Known pockets are colored in purple and the new pockets are shown in blue. It can be seen that the clustering results in well-defined groups, many containing both known and new. Edges are drawn between two nodes (pockets) only if they share a PMSmax ≥ 0.7. (C1–C3) Individual graphs for clusters that contain very similar sites coming from different fold families. Nodes in each of the three panels are colored based on the folds, and it can be seen that there are at least 16 different folds in C1, for example. (D1–D3) Example alignments for a pair of pockets from each cluster. Pocket residues belonging to two different proteins in a pair that are similar are shown in red and blue respectively. D1 shows a pair of pockets in a particular site type, which have similarity to a menaquinone binding pocket, and similarly in D2 and D3, a pair of pockets from two different site types showing similarity to NAD and heme binding respectively. The ligand is shown in green in each panel. The chemical and geometrical similarities in the pockets are seen to be quite high. Structure 2018 26, 499-512.e2DOI: (10.1016/j.str.2018.02.001) Copyright © 2018 Elsevier Ltd Terms and Conditions

Figure 5 Aspects of Structural Annotation A schematic representation of fold-type to site-type to ligand-type association derived from the pocketome analysis. The six different categories discussed are (A) one fold, one site type, one ligand type; (B) one fold, one site type, multiple ligand types; (C) multiple folds, one site type, one ligand type; (D) multiple folds, one site type, multiple ligand types; (E) one fold,- multiple site types, one ligand type; and (F) one fold, multiple site types, multiple ligand types. The number of these association types seen in the pocketome is also indicated. Structure 2018 26, 499-512.e2DOI: (10.1016/j.str.2018.02.001) Copyright © 2018 Elsevier Ltd Terms and Conditions

Figure 6 Analysis of Allosteric Proteins (A1–A6) Pairwise alignments suggesting new modulators to bind known allosteric pockets (blue) because of binding site similarity between the known allosteric pockets (blue) and the known binding site of the new modulator in its original protein (red). The possible modulator ligand is shown in ball and stick colored by atom type in all panels. The ligands in (A1–A3) are drugs/drug-like molecules. (B1–B4) Druggable proteins and identification of drug-target association using pockets from the augmented pocketome. Binding site similarities are shown for four example proteins (which are non-obvious at the sequence level), aldehyde dehydrogenase (1O9J, blue) with trimetrexate-TMQ (3HBB, red); beta1-adrenoceptor (4BVN, blue) with diflunisal-1FL (2BXE, red); isocitrate dehydrogenase (4HCX, blue) with mevastatin-114 (1HW8, red); and orexin receptor (4S0V, blue) with alprazolam-08H (3U5J, red). Here again, the drug ligands in each panel are shown in ball and stick model colored by atom type. Structure 2018 26, 499-512.e2DOI: (10.1016/j.str.2018.02.001) Copyright © 2018 Elsevier Ltd Terms and Conditions