Poster 9: Construction and Use of a Fragment Virtual Screening LibraryRichard Hall1
|1Astex Pharmaceuticals, 436 Cambridge Science Park, Milton Road, Cambridge, CB4 0QA|
|Fragment based drug discovery has emerged as an alternative to high throughput screening in drug discovery. The screening of focused libraries of small, functionally simple, molecules has been shown to be successful. The B-Raf(V600E) inhibitor, Vemurafenib (Plexxikon) is now approved and several other fragment derived compounds are progressing through the clinic.|
Astex are one of the world leaders in fragment based drug discovery. Our fragment libraries have evolved over the lifetime of the company and we have been able to use our database of crystal structures and screening data to guide this process. Our fragment libraries are screened using a number of techniques. As well as X-ray crystallography and other biophysical methods, we have a wealth of experience in docking and virtual screening. Because our projects start with one or more crystal structures, we are able to complement a primary screening campaign with the use of 3D computational methods that probe the protein active site and provide ideas for new compounds for purchase or synthesis.
We have recently generated a virtual screening library of thirty thousand fragments that are available for purchase from preferred suppliers and that can be screened at the start of a drug discovery project. We here describe the chemoinformatic effort that went into choosing the fragments for the library. Some of the points for consideration are:
How does one filter the available compounds to remove fragments that have undesirable properties?
How reputable is the supplier - we are interested in fragments that can be purchased in a timely fashion?
How similar are the fragments to one another - can we cluster the fragments in a useful way?
How do we choose fragments from the clusters that are generated?
Docking and scoring of fragments is a challenging problem. We shall use our newly constructed library in a number of virtual screening campaigns across a diverse set of protein targets to generate an ensemble of scores for each fragment. We shall investigate ways in which these reference energies can be used to normalise the score of fragments and what benefit this may provide over a simple ranked scoring approach.