Panel docking of small-molecule libraries — Prospects to improve efficiency of lead compound discovery

Abstract

Computational docking as a means to prioritise small molecules in drug discovery projects remains a highly popular in silico screening approach. Contemporary docking approaches without experimental parametrisation can reliably differentiate active and inactive chemotypes in a protein binding site, but the absence of a correlation between the score of a predicted binding pose and the biological activity of the molecule presents a clear limitation.

Several novel or improved computational approaches have been developed in the recent past to aid in screening and profiling of small-molecule ligands for drug discovery, but also more broadly in developing conceptual relationships between different protein targets by chemical probing. Among those new methodologies is a strategy known as inverse virtual screening, which involves the docking of a compound into different protein structures. In the present article, we review the different computational screening methodologies that employ docking of atomic models, and, by means of a case study, present an approach that expands the inverse virtual screening concept.

By computationally screening a reasonably sized library of 1235 compounds against a panel of 48 mostly human kinases, we have been able to identify five groups of putative lead compounds with substantial diversity when compared to each other. One representative of each of the five groups was synthesised, and tested in kinase inhibition assays, yielding two compounds with micro-molar inhibition in five human kinases.

This highly economic and cost-effective methodology holds great promise for drug discovery projects, especially in cases where a group of target proteins share high structural similarity in their binding sites.

Introduction

Because of the increasing importance placed on the pursuit of innovative chemical starting points (leads) in contemporary small-molecule drug discovery (Leeson and Springthorpe, 2007, Oprea, 2002), the challenge to identify non-promiscuous kinase inhibitors is made even more daunting on account of the limited number of known chemotypes. The search for new lead structures in contemporary drug discovery typically entails the high-throughput screening of megalibraries of small organic molecules (molecules with a molecular mass < 500 Da) against modified cell lines or isolated recombinantly produced target proteins to identify novel modulators (Macarron et al., 2011). While in the field of protein kinases, for example, many ATP-competitive inhibitors were indeed discovered via this strategy, high-throughput screening has lost much of its initial momentum as most of the scaffolds present in existing libraries have now been identified (Li et al., 2004). Another limitation associated with the high-throughput screening of compound libraries (especially against a comprehensive panel of protein kinases) is that the process can be time and resource intensive (Camp et al., 2012). Because of these drawbacks, predictive methods have increasingly been employed to identify new modulators that could potentially be developed into drugs.

In the past 15 years, a variety of different informatics-based and computational methods have been developed to aid in screening and profiling of small-molecule ligands for drug discovery (for a review see (Ekins et al., 2007)). A relatively new strategy known as inverse virtual screening (IVS) involves the docking of a particular compound into different protein structures described in protein databases (Chen and Zhi, 2001, Hui-fang et al., 2010, Lauro et al., 2011, Paul et al., 2004). Here, we present a brief overview of the different computational screening methodologies based on docking, and show that the resulting “panel docking” scores potentially facilitate the identification and analysis of promising modulators through statistical analysis of predicted binding to more than one target protein.