Challenges and advances in structure-based virtual screening
2013; Future Science Ltd; Volume: 6; Issue: 1 Linguagem: Inglês
10.4155/fmc.13.186
ISSN1756-8927
Autores Tópico(s)Bioinformatics and Genomic Networks
ResumoFuture Medicinal ChemistryVol. 6, No. 1 EditorialFree AccessChallenges and advances in structure-based virtual screeningElizabeth YurievElizabeth YurievMedicinal Chemistry, Monash Institute of Pharmaceutical Sciences, Monash University (Parkville Campus), 381 Royal Parade, Parkville, VIC 3052 Australia. Published Online:23 Dec 2013https://doi.org/10.4155/fmc.13.186AboutSectionsPDF/EPUB ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinkedInRedditEmail Keywords: computational efficiencyhigh-performance computingstructure-based virtual screeningvirtual decoysIn the 1980s, rational drug design was proclaimed as a way to discover drugs that would cut down on the amount of experimentation required. A lot of effective science resulted from this approach, but designing compounds one at a time, especially on the then slow computers, did not fill the pipeline. In the 1990s, the pairing of combinatorial chemistry and high-throughput screening (HTS) was hailed as a panacea because it did not require rational knowledge about the mode of drug action and relied mostly on testing as many compounds as possible. While absolutely essential in modern drug discovery, HTS suffered from being expensive and, therefore, not as productive as initially hoped. After all, making the haystack bigger does not help finding that proverbial needle. Virtual screening, and in particular structure-based virtual screening (SBVS), evolved as an adaptive response [1]. It inherited the best of both approaches: a sensible consideration of drug–target binding came from rational drug design and the screening methodology mirrored that of HTS. Today, SBVS technology is the major beneficiary of two key areas of scientific advancement. The first is the progress in target identification, through genomics, proteomics, x-ray crystallography and NMR. The second is the development of computational methods, through advances in hardware and algorithms. As a result, SBVS has become an integral part of the strategies that pharmaceutical companies and academic laboratories employ when undertaking drug-discovery research. These endeavors led to an increasing number of successful SBVS campaigns that have identified useful drug leads (see [1] for exemplar studies).However, importantly and somewhat intriguingly, SBVS is not perfect. As SBVS is based on computational docking, it suffers from all the challenges faced by docking and scoring. Specifically, it needs to account for the conceptually important yet difficult-to-compute notions of receptor flexibility, solvation and entropic contributions to binding. New methodological advances in docking are constantly addressing these challenges [2,3]. While these advances move the field forward, the question still remains: given the shortcomings of scoring functions and the magnitude of the problem (having to dock millions of ligands into any given target or several possible targets), why does SBVS actually 'work'? The answer is simple: computational screening is an enrichment process. Accurately calculated binding energies and scores are not necessarily required for meaningful compound selection. Finding active compounds in the shortlist is, however, critically important. Appropriate selection strategies, therefore, compensate for methodological shortcomings, while deselection of inappropriate compounds reduces the risk of taking a non-promising candidate through a drug-discovery campaign. Therefore, two critical drivers in SBVS protocol development are the increase in the success rate for finding novel actives and the improvement of computational efficiency.One of the important questions facing practitioners of SBVS is 'How to measure SBVS success?'. This issue has been widely discussed in the field (for an example see [4]). Most commonly used measures of success in retrospective SBVS include enrichment factors and the area under the receiver operating characteristic curve (ROC AUC). While enrichment plots and enrichment factors are still routinely used for measuring virtual screening performance (for example [5]), they are not ideal. ROC curves are superior to enrichment plots as they reflect the selection of actives as well as the non-selection of decoys [4,6]. ROC AUC gives an indication of the total number of compounds successfully docked into the model and is interpreted as the probability that a randomly chosen active has a higher score than a randomly chosen inactive. Several metrics, such as normalized square root AUC [7] and LogAUC [8] have also been developed to focus on early, rather than overall, enrichment.Along with metrics to judge SBVS success rates in retrospective evaluations, a significant effort has gone into the development of appropriate decoy sets to use in these studies. Among the most common decoy sets used are Directory of Useful Decoys [101], Schrödinger decoy set [102] and miscellaneous filtered versions of ZINC [103]. These libraries contain commercially available compounds, which is a useful feature for prospective screening, but is not necessary in retrospective method evaluation. This feature even leads to some limitations as these decoy sets span a small, synthetically feasible subset of molecular space and are restricted in physicochemical similarity compared with actives. For retrospective screening, decoys do not need to be 'real'. Virtual decoys should be chemically possible but not necessarily synthetically feasible. Their advantage is that they could be designed and physicochemically matched for any active. Using the virtual decoy sets and demanding evaluation kits for objective in silico screening [9,10], it was demonstrated that it is possible to benchmark scoring functions and assess their robustness as well as advantages and limitations. Another important direction is the development of protein-specific decoys. Retrospective screening with such challenging decoys allows more vigorous SBVS method validation. For example, a G protein-coupled receptor decoy library with 39 decoy molecules selected for each G protein-coupled receptor ligand was developed [11].The criteria that contribute to the success of SBVS may also be established by looking at a selection of prospective studies. Ripphausen et al. systematically evaluated the state-of-the-art in SBVS by surveying 279 prospective studies, published during July 2011 [12]. They observed that high resolution of structural targets and sophistication of scoring functions were not actually decisive factors for the success of SBVS. Instead, scientific expertise, chemical intuition and subjective compound rankings played a more important role in compound selection for testing.A similar survey of virtual screening studies published between 2007 and 2011 was recently performed by Zhu et al.[13]. They also addressed the issue of compound selection for testing and, using their observations, strongly recommended using ligand efficiency (LE) for both hit identification and hit optimization stages. In particular, they endorsed target LE, which is an adjusted LE based on the molecular size of the screened compounds. Another consideration for compound selection for testing is chemotype novelty. Metrics such as 'cluster averaging' [14], where the contribution of each active to the score is proportional to the number of other actives of the same chemotype, are useful when performing SBVS for scaffold-hopping purposes.Undoubtedly, the success of SBVS depends on computational efficiency. While significantly 'cheaper' than HTS and well supported by increasing availability of high-performance computing (HPC) resources, the significant cost of virtual screening is still computer time. To address the issue of reducing the computational cost of virtual screening, Skone et al. adopted the 'lazy evaluation' principle from computer science: 'a calculation that makes no contribution to the final outcome should be avoided'. In a study fittingly entitled 'Knowing when to give up: early rejection stratagems in ligand docking' [15], they were able to reduce the run times of screening without any significant impairment to docking outcomes. This principle should be implemented in a wide range of docking programs.Other approaches to improving computational efficiency of SBVS have recently included increasing automation of the process as well as exploitation of grid/cloud resources [104]. In many cases, docking is still manually intensive and requires expert handling and decision making. To make SBVS truly useful for medicinal chemists, it should become fully automatable, with the use of integrated computational platforms such as the DOCK Blaster server [16,105]. It must be noted that the DOCK Blaster server has initially delivered good pose fidelity but achieved enrichment only in 25–40% of cases [16]. With a caveat that these results are relatively poor, especially when compared with expert studies, DOCK Blaster clearly allows the exploitation of the increasing amount of structural data for drug discovery.The rise of HPC has recently led to improvements in data management in parallel applications in SBVS. The Docking@Home project allowed the distribution of SBDV calculations among volunteer/general public computers [17,106], while the Chemomentum computing environment [18] has combined paradigms of grid computing and collaborative research. In addition to automation and utilization of distributed resources for data management, docking algorithm developers have recently been creating SBVS tools, which can run on multi-core systems and grid architectures due to their parallel design; see references for examples of such programs [2,3].Future improvements in virtual screening success should come from several directions. Thanks to progress in HPC, we should soon be able to carry out enormous virtual screens within practical time spans. We also now have good tools for evaluating SBVS protocols and predicting whether they are expected to be useful for prospective screening. The challenge that remains is our ability to accurately predict binding affinities of drug candidates. In order to achieve this goal, an improvement in accounting for protein flexibility, solvation and entropic effects is required [2,3]. Whether it will be achieved via machine-learning approaches and generalized/universal scoring functions or by developing protein-specific/targeted scoring functions remains to be seen [2,3,19].Financial & competing interests disclosureThe author has no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.No writing assistance was utilized in the production of this manuscript.References1 Kar S, Roy K. How far can virtual screening take us in drug discovery? Expert Opin. Drug. Discov.8(3),245–261 (2013).Crossref, Medline, CAS, Google Scholar2 Yuriev E, Ramsland PA. Latest developments in molecular docking: 2010–2011 in review. J. Mol. Recognit.26(5),215–239 (2013).Crossref, Medline, CAS, Google Scholar3 Yuriev E, Agostino M, Ramsland PA. Challenges and advances in computational docking: 2009 in review. J. Mol. Recognit.24(2),149–164 (2011).Crossref, Medline, CAS, Google Scholar4 Nicholls A. What do we know and when do we know it? J. Comput. Aided Mol. Des.22(3–4),239–255 (2008).Crossref, Medline, CAS, Google Scholar5 Halgren TA, Murphy RB, Friesner RA et al. Glide: a new approach for rapid, accurate docking and scoring. 2. Enrichment factors in database screening. J. Med. Chem.47(7),1750–1759 (2004).Crossref, Medline, CAS, Google Scholar6 Hawkins PCD, Warren GL, Skillman AG, Nicholls A. How to do an evaluation: pitfalls and traps. J. Comput. Aided Mol. Des.22(3–4),179–190 (2008).Crossref, Medline, CAS, Google Scholar7 Katritch V, Rueda M, Lam PC, Yeager M, Abagyan R. GPCR 3D homology models for ligand screening: lessons learned from blind predictions of adenosine A2a receptor complex. Proteins78(1),197–211 (2010).Crossref, Medline, CAS, Google Scholar8 Mysinger MM, Shoichet BK. Rapid context-dependent ligand desolvation in molecular docking. J. Chem. Inf. Model.50(9),1561–1573 (2010).Crossref, Medline, CAS, Google Scholar9 Wallach I, Lilien R. Virtual decoy sets for molecular docking benchmarks. J. Chem. Inf. Model.51(2),196–202 (2011).Crossref, Medline, CAS, Google Scholar10 Vogel SM, Bauer MR, Boeckler FM. DEKOIS: demanding evaluation kits for objective in silico screening – a versatile tool for benchmarking docking programs and scoring functions. J. Chem. Inf. Model.51(10),2650–2665 (2011).Crossref, Medline, CAS, Google Scholar11 Gatica EA, Cavasotto CN. Ligand and decoy sets for docking to G protein-coupled receptors. J. Chem. Inf. Model.52(1),1–6 (2012).Crossref, Medline, CAS, Google Scholar12 Ripphausen P, Stumpfe D, Bajorath J. Analysis of structure-based virtual screening studies and characterization of identified active compounds. Future Med. Chem.4(5),603–613 (2012).Link, CAS, Google Scholar13 Zhu T, Cao S, Su PC et al. Hit identification and optimization in virtual screening: practical recommendations based on a critical literature analysis. J. Med. Chem.56(17),6560–6572 (2013).Crossref, Medline, CAS, Google Scholar14 Mackey MD, Melville JL. Better than random? The chemotype enrichment problem. J. Chem. Inf. Model.49(5),1154–1162 (2009).Crossref, Medline, CAS, Google Scholar15 Skone G, Voiculescu I, Cameron S. Knowing when to give up: early-rejection stratagems in ligand docking. J. Comput. Aided Mol. Des.23(10),715–724 (2009).Crossref, Medline, CAS, Google Scholar16 Irwin JJ, Shoichet BK, Mysinger MM et al. Automated docking screens: a feasibility study. J. Med. Chem.52(18),5712–5720 (2009).Crossref, Medline, CAS, Google Scholar17 Taufer M, Armen R, Chen J, Teller P, Brooks C. Computational multiscale modeling in protein-ligand docking. IEEE Eng. Med. Biol. Mag.28(2),58–69 (2009).Crossref, Medline, Google Scholar18 Garcia-Sosa AT, Sild S, Maran U. Docking and virtual screening using distributed grid technology. QSAR Comb. Sci.28(8),815–821 (2009).Crossref, CAS, Google Scholar19 Ross GA, Morris GM, Biggin PC. One size does not fit all: the limits of structure-based models in drug discovery. J. Chem. Theory Comput.9(9),4266–4274 (2013).Crossref, Medline, CAS, Google Scholar101 A directory of useful decoys. http://dud.docking.orgGoogle Scholar102 Schrödinger. www.schrodinger.comGoogle Scholar103 ZINC. http://zinc.docking.orgGoogle Scholar104 Cloud computing: a drug discovery game changer? www.inhibox.com/sites/default/files/CloudComputingADrugDiscoveryGameChanger.pdfGoogle Scholar105 DOCK Blaster. http://blaster.docking.orgGoogle Scholar106 Docking@Home. http://docking.cis.udel.eduGoogle ScholarFiguresReferencesRelatedDetailsCited ByLigand and structure based hierarchical virtual screening cascade for finding novel epidermal growth factor receptor inhibitors17 October 2023 | Chemical Biology & Drug Design, Vol. 12Computational investigation of phytochemicals from Abrus precatorius seeds as modulators of peroxisome proliferator-activated receptor gamma (PPARγ)30 June 2022 | Journal of Biomolecular Structure and Dynamics, Vol. 41, No. 12Small Molecular Drug Screening Based on Clinical Therapeutic Effect27 July 2022 | Molecules, Vol. 27, No. 15Generating property-matched decoy molecules using deep learning3 February 2021 | Bioinformatics, Vol. 37, No. 15An Integrative in silico Drug Repurposing Approach for Identification of Potential Inhibitors of SARS‐CoV‐2 Main Protease30 March 2021 | Molecular Informatics, Vol. 40, No. 5Development and Evaluation of MM/GBSA Based on a Variable Dielectric GB Model for Predicting Protein–Ligand Binding Affinities16 March 2020 | Journal of Chemical Information and Modeling, Vol. 60, No. 11Structural binding perspectives of a major tobacco alkaloid, nicotine, and its metabolite cotinine with sex‐steroid nuclear receptors28 April 2020 | Journal of Applied Toxicology, Vol. 40, No. 10Drug screening with the Autodock Vina on a set of kinases without experimentally established structuresVirtual Screening in the Cloud: How Big Is Big Enough?4 November 2019 | Journal of Chemical Information and Modeling, Vol. 60, No. 9Structure investigation, enrichment analysis and structure-based repurposing of FDA-approved drugs as inhibitors of BET-BRD417 November 2018 | Journal of Biomolecular Structure and Dynamics, Vol. 37, No. 12G-quadruplex virtual drug screening: A reviewBiochimie, Vol. 152Protein structure prediction provides comparable performance to crystallographic structures in docking-based virtual screeningMethods, Vol. 71Allosteric mechanisms of nuclear receptors: insights from computational simulationsMolecular and Cellular Endocrinology, Vol. 393, No. 1-2 Vol. 6, No. 1 STAY CONNECTED Metrics History Published online 23 December 2013 Published in print January 2014 Information© Future Science LtdKeywordscomputational efficiencyhigh-performance computingstructure-based virtual screeningvirtual decoysFinancial & competing interests disclosureThe author has no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.No writing assistance was utilized in the production of this manuscript.PDF download
Referência(s)