High-throughput screening (HTS) is a rapid but broad measure of compound activity, frequently used at the start of projects. The assays are typically run at a single concentration for a single measure of percent inhibition, and they often result in false negatives.
Virtual screening (VS) can be used as soon as there is data available about an active compound, either from wet screening or from the literature. VS is computationally expensive, but a recent port of VS software to use GPU hardware means that this method is now accessible to most research facilities.
Enriching your set of candidate compounds
VS is a simple process that can be implemented as part of a standard assay cascade. For example, a company may choose to progress 1000 compounds from a wet screen and add in an additional 100 compounds identified by the virtual screen. The additional compounds may identify classes of actives that were not identified by the original HTS run, enriching the list of possibilities available for progression into hit-to-lead.
Cost-effective screening
When active compounds are already known from the literature, VS can even replace HTS. Running a VS on a large, initial data set is orders of magnitude cheaper than running a HTS on the same compounds.
The VS can be used to dramatically reduce the number of compounds that need to be screened, resulting in large cost savings. Done well, VS results in a similar dataset to HTS at a fraction of the cost.
Ligand- or protein-based virtual screening
There are two main approaches to VS: ligand-based VS and protein-structure based VS. The difference is whether the protein or the ligand is taken as the template against which compounds are screened.
Each approach has strengths, and the results from either method can be comparable and often highly complementary. However, the great advantage of ligand-based VS is that it can be used when there is no information available about the protein structure. This is likely to be the case at the start of projects when HTS is being considered.
Ligand-based VS uses a template molecule that is known to bind to the target, or a pharmacophore derived from this template. It compares this to each molecule in the database in order to find molecules that are in some way similar to the known actives. The principle that similar molecules share similar activity results in a final data set that contains all the compounds likely to be active at the target of interest.
Virtual screening for 3D similarity
Some VS methods focus mainly on the 2D structure of compounds. However, since ligand interactions are based on the shape and electrostatic character of molecules, compounds that share biological activity are likely to have similarities that go well beyond their chemical structure. 3D VS is therefore far more likely to identify novel active molecules than 2D VS but at the cost of increased computational effort.
The case study “Virtual Screening for Diverse New Leads for 11β-HSD-1” gives an example of how 3D VS can be used to identify diverse new active chemotypes. 1
GPUs make routine 3D virtual screening viable
3D VS involves a very large amount of data and until recently has required a significant hardware investment (around $100,000) to get results in a useful timescale.
Graphical processing units (GPUs), developed in the gaming industry, use massively parallel architecture to speed up computationally intense applications. The VS software Blaze was recently ported to GPUs using OpenCL. Tests show that using BlazeGPU achieves a 40x speed-up while losing nothing in scientific accuracy.2
This speed up can translate to faster calculations, larger data sets or lower hardware costs. Four commodity graphics cards can produce the same results as a 150-node CPU cluster, making it possible to run a large scale VS on hardware costing as little as $10,000. Organisations can now afford to run VS alongside every HTS as a standard method of minimizing false negatives and identifying the best possible set of candidate compounds for progression to secondary screening.
References
- Slater M, Cheeseright T, Vinter A. Virtual Screening for Diverse New Leads for 11β-HSD-1. http://www.cresset-group.com/wp-content/uploads/2013/11/Virtual-screening-resulting-in-diverse-new-leads.pdf Accessed December 20, 2013.
- Krige S, Mackey M, McIntosh-Smith S, Sessions R. Porting a commercial application to OpenCL: a case study. Presented at: The 1st International Workshop on OpenCL (IWOCL); May 2013; Atlanta, Georgia.
