Encoding protein-ligand interaction patterns in fingerprints and graphs

sc-PDB: a 3D-database of ligandable binding sites--10 years on

The sc-PDB database (available at http://bioinfo-pharma.u-strasbg.fr/scPDB/) is a comprehensive and up-to-date selection of ligandable binding sites of the Protein Data Bank. Sites are defined from complexes between a protein and a pharmacological ligand. The database provides the all-atom description of the protein, its ligand, their binding site and their binding mode. Currently, the sc-PDB archive registers 9283 binding sites from 3678 unique proteins and 5608 unique ligands. The sc-PDB database was publicly launched in 2004 with the aim of providing structure files suitable for computational approaches to drug design, such as docking. During the last 10 years we have improved and standardized the processes for (i) identifying binding sites, (ii) correcting structures, (iii) annotating protein function and ligand properties and (iv) characterizing their binding mode. This paper presents the latest enhancements in the database, specifically pertaining to the representation of molecular interaction and to the similarity between ligand/protein binding patterns. The new website puts emphasis in pictorial analysis of data.

Beware of machine learning-based scoring functions-on the danger of developing black boxes

Training machine learning algorithms with protein-ligand descriptors has recently gained considerable attention to predict binding constants from atomic coordinates. Starting from a series of recent reports stating the advantages of this approach over empirical scoring functions, we could indeed reproduce the claimed superiority of Random Forest and Support Vector Machine-based scoring functions to predict experimental binding constants from protein-ligand X-ray structures of the PDBBind dataset. Strikingly, these scoring functions, trained on simple protein-ligand element-element distance counts, were almost unable to enrich virtual screening hit lists in true actives upon docking experiments of 10 reference DUD-E datasets; this is a a feature that, however, has been verified for an a priori less-accurate empirical scoring function (Surflex-Dock). By systematically varying ligand poses from true X-ray coordinates, we show that the Surflex-Dock scoring function is logically sensitive to the quality of docking poses. Conversely, our machine-learning based scoring functions are totally insensitive to docking poses (up to 10 Å root-mean square deviations) and just describe atomic element counts. This report does not disqualify using machine learning algorithms to design scoring functions. Protein-ligand element-element distance counts should however be used with extreme caution and only applied in a meaningful way. To avoid developing novel but meaningless scoring functions, we propose that two additional benchmarking tests must be systematically done when developing novel scoring functions: (i) sensitivity to docking pose accuracy, and (ii) ability to enrich hit lists in true actives upon structure-based (docking, receptor-ligand pharmacophore) virtual screening of reference datasets.

Polypharmacology rescored: protein-ligand interaction profiles for remote binding site similarity assessment

Detection of remote binding site similarity in proteins plays an important role for drug repositioning and off-target effect prediction. Various non-covalent interactions such as hydrogen bonds and van-der-Waals forces drive ligands' molecular recognition by binding sites in proteins. The increasing amount of available structures of protein-small molecule complexes enabled the development of comparative approaches. Several methods have been developed to characterize and compare protein-ligand interaction patterns. Usually implemented as fingerprints, these are mainly used for post processing docking scores and (off-)target prediction. In the latter application, interaction profiles detect similarities in the bound interactions of different ligands and thus identify essential interactions between a protein and its small molecule ligands. Interaction pattern similarity correlates with binding site similarity and is thus contributing to a higher precision in binding site similarity assessment of proteins with distinct global structure. This renders it valuable for existing drug repositioning approaches in structural bioinformatics. Current methods to characterize and compare structure-based interaction patterns - both for protein-small-molecule and protein-protein interactions - as well as their potential in target prediction will be reviewed in this article. The question of how the set of interaction types, flexibility or water-mediated interactions, influence the comparison of interaction patterns will be discussed. Due to the wealth of protein-ligand structures available today, predicted targets can be ranked by comparing their ligand interaction pattern to patterns of the known target. Such knowledge-based methods offer high precision in comparison to methods comparing whole binding sites based on shape and amino acid physicochemical similarity.