Troubleshooting computational methods in drug discovery

Abstract

Computational approaches for drug discovery such as ligand-based and structure-based methods, are increasingly seen as an efficient approach for lead discovery as well as providing insights on absorption, distribution, metabolism, excretion and toxicity (ADME/Tox). What is perhaps less well known and widely described are the limitations of the different technologies. We have therefore presented a troubleshooting approach to QSAR, homology modeling, docking as well as hybrid methods. If such computational or cheminformatics methods are to become more widely used by non-experts it is critical that such limitations are brought to the user's attention and addressed during their workflows. This could improve the quality of the models and results that are obtained.

Introduction

Lead generation and avoiding late stage failures of lead compounds are the two major bottlenecks in the drug discovery process. The problems associated with lead discovery starts with the escalated costs associated with lead generation against a validated target. Further attempts to utilize combinatorial and high throughput screening has lead to a marginal increase in success which had prompted pharmaceutical and biotech industries to question the cost effectiveness of these technologies (Macarron, 2006). Similarly, problems associated with late stage failures of potent lead compounds due to undesirable physicochemical properties has lead to a major shift in the drug discovery protocols over the past decade to evaluate lead compounds for drug like properties very early on in the discovery process using computational prediction methods based on statistical techniques that utilize existing experimental knowledge about physicochemical properties (Ekins, Ring, Grace, McRobie-Belle & Wrighton, 2000c). Virtual screening techniques have been integrated into the design protocol thereby reducing the burden on experimentally screening millions of compounds to identify lead molecules and hence provide very cost effective methods to solve the lead generation issues (Oprea & Matter, 2004).

In silico methods can be broadly listed as those that can be applied to study a) targets such as protein and DNA sequences, genomes, networks or pathways and even tissues and organs such as virtual liver b) small molecules such as those that do not contain the amino acids or nucleic acids and c) complexes of small molecules with target proteins or DNA or RNA. Based on these classifications, in silico techniques have been utilized for analyzing target and small molecule interactions and networks (Bajorath, 2008, Hopkins, 2007, Oprea et al., 2007, Young et al., 2008), quantitative structure activity relationships (Dimitrov et al., 2005, Dudek et al., 2006), analyzing similarity between either small molecules or targets (Lemmen & Lengauer, 2000, Melville et al., 2009, Schaffer et al., 2001, Sharan & Ideker, 2006, Williams & Schreyer, 2009), pharmacophore based modeling and screening (Chang et al., 2006c, Khedkar et al., 2007, Sun, 2008), homology modeling (Hillisch et al., 2004, Qu et al., 2009, Xu et al., 2000), molecular dynamics simulation protocols (Amaro & Li, 2010, Morra et al., 2010, Okimoto et al., 2009), data mining of genomes or small molecule databases (Feng et al., 2009, Marechal, 2008, Wagner & Clemons, 2009), network or pathway analysis (Hendriks et al., 2008, Hopkins, 2008), machine learning techniques etc. (Chekmarev et al., 2009a, Kortagere et al., 2009b, Li et al., 2007, Melville et al., 2009). Thus modern drug discovery is a complex, iterative process involving several levels of screening the wealth of experimental data available from whole genome sequencing projects to those derived from public and private databases of chemicals such as PubChem (Sayers et al., 2009), Zinc (Irwin & Shoichet, 2005) and ChemSpider (Williams, 2008). Extracting all the information at the required level of sophistication has been made possible only with the exceptional bioinformatics and computational tools. Although computational methods have been designed for a variety of applications listed above, this study concentrates on an overview of limitations and ways to troubleshoot them for the in silico techniques relevant to virtual screening and absorption, distribution, metabolism, excretion and toxicity (ADME/Tox) filtering.