Binding site matching in rational drug design: algorithms and applications

Abstract

Interactions between proteins and small molecules are critical for biological functions. These interactions often occur in small cavities within protein structures, known as ligand-binding pockets. Understanding the physicochemical qualities of binding pockets is essential to improve not only our basic knowledge of biological systems, but also drug development procedures. In order to quantify similarities among pockets in terms of their geometries and chemical properties, either bound ligands can be compared to one another or binding sites can be matched directly. Both perspectives routinely take advantage of computational methods including various techniques to represent and compare small molecules as well as local protein structures. In this review, we survey 12 tools widely used to match pockets. These methods are divided into five categories based on the algorithm implemented to construct binding-site alignments. In addition to the comprehensive analysis of their algorithms, test sets and the performance of each method are described. We also discuss general pharmacological applications of computational pocket matching in drug repurposing, polypharmacology and side effects. Reflecting on the importance of these techniques in drug discovery, in the end, we elaborate on the development of more accurate meta-predictors, the incorporation of protein flexibility and the integration of powerful artificial intelligence technologies such as deep learning.

Keywords: drug repositioning; drug side effects; off-targets; pocket alignment; pocket matching; polypharmacology.

Figure 1

Various pocket representation methods. An ATP molecule colored by atom type (C – peach, N – blue, O – red, P – orange) bound to the pocket of protein kinase C iota type is shown in the center with four selected binding residues, G252, Y256, A272 and D387 represented as green, cyan, blue and red sticks, respectively. (A) All-atom pocket representation, in which individual atoms are depicted by small spheres colored by atom type (N – blue, O – red, C – green/cyan/blue/red). (B) Sets of functional centers of selected microenvironments, hypothetical Cβ (green), Cβ (blue), aromatic (cyan) and negatively charged (red). Functional centers are represented by small solid spheres, whereas large, transparent spheres correspond to the local microenvironments. (C) A mixed representation of pocket residues employing backbone atoms (small spheres colored as in A) and the side-chain centroids (large, gray spheres). (D) A coarse-grained model based on Cα atoms, which are depicted by solid spheres colored according to residues in the center image.

Figure 2

Similar binding sites in two globally unrelated proteins. ATP molecules bound to (A) human protein kinase C iota type, PKC-iota (purple) and (B) N5-carboxyaminoimidazole ribonucleotide synthetase, purK from E. coli (gold). The Cα atoms of four selected pocket residues in each structure are represented by solid spheres and labeled. (C) The superposition of PKC-iota and purK based on ATP molecules. Equivalent residue pairs indicated by double-headed arrows form the local alignment of two ATP-binding sites.

Figure 3

Different algorithms to align ligand-binding sites. Three types of techniques are presented, (A and B) the clique detection, (C and D) the assignment method and (E) the geometric sorting, for a pair of ATP-binding sites in PKC-iota (purple) and purK (gold) whose molecular structures are shown in Figure 2. (A) Graph representations of both pockets with binding residues depicted as vertices connected by dotted edges indicating close positions in the structure. (B) A modular product of the two graphs displayed on the top (PKC-iota) and the left side (purK). Three instances of a four-node subgraph are highlighted by thick solid lines with the maximum clique colored purple and gold. (C) An assignment matrix constructed for binding sites in PKC-iota (rows) and purK (columns) populated with pairwise residue-based scores. The optimal alignment obtained by solving the linear sum assignment problem (LSAP) is marked by solid boxes. (D) A bipartite graph showing all the possible one-to-one alignments between PKC-iota and purK binding residues with the optimal alignment found in C marked by solid lines. (E) Sorting of PKC-iota/purK residue pairs according to the assigned scores. The pocket alignment is constructed by eliminating those pairs having a residue that appears in a higher-ranked pair.

Figure 4

Diagram of selected applications of pocket matching algorithms. A blue structure in the center is the primary target for drug d. Binding sites in other proteins that are similar to the d-binding pocket in the primary target are colored red. (A) Repurposing of d to another protein colored green implicated in a different disease than the primary target is associated with. (B) An illustration of the concept of pocket-based polypharmacology. If the primary target is part of a disease-related pathway involving other proteins with similar binding sites, marked with asterisks, then d or its derivatives are candidates for the polypharmacological action on that pathway. (C) An analysis of drug side effects caused by off-target binding. A purple protein identified to have a similar pocket to that in the blue structure is a potential off-target for d.