Fragment-based screening, what can we learn from published hits?

Fragment-based screening, what can we learn from published hits? was presented by Dr Chris Swain, Cambridge MedChem Consulting at the Cresset European User Group Meeting 2015.

Chris Swain has compiled a database of fragments from the literature. His goal in doing this was to design better ligands, but he has found that having a dataset of fragments bound to proteins allows you to ask other interesting questions.

Fragment-based screening is a technology for finding starting points for drug discovery. Its popularity is based on the fact that it is much cheaper than traditional methods. Also, fragment space is much smaller than chemical space, therefore a fragment screen can cover much of fragment space with only 2000 molecules.

Designing a fragment library

Astex proposed the ‘rule of 3’ when selecting fragments to go into the library. This rule states that you should choose fragments that have the following range of properties:

However, solubility is in many ways more important. All of your insoluble compounds tend to cause artefacts. Many people have tried to design better libraries choosing fragments based on known drugs, PDB ligands, natural products, or enhanced 3D structure. The literature data, of which there is a lot, can be used to design better fragment libraries.

Chris designed a database of fragment hits taken from the literature. He also included additional data in the database including the screening technology, the target and Uniprot ID, the affinity and how it was measured, the PDB code and the target type and class. He also calculated a lot of physicochemical properties, the functional groups and a cluster analysis.

Finding the data in the literature is getting more and more difficult. As fragment screening becomes more general, it’s not always mentioned in the title.

As of 1 May 2015, the database covers:

• 213 Publications
• 1,036 Published hits
• 152 Different targets
• 23 Detection technologies.

Information that can be mined from the fragments database

One of the things you can do with the database is to look at where the screening hits came from. Maybridge are far and away the highest supplier. This could well be because they were the first to include solubility in their database. The other thing to consider is clustering. Most of them turn out to be singletons, very few are clustered. A lot of the fingerprints are sparsely populated, so it’s not the best way to measure diversity. Cresset might be a better way of measuring diversity. The vast majority of the fragments are aromatic, many of which are heterocycles. Approximately 400 of the compounds have an ionisable group. It seems that this may be important as an interaction. 26 contain nitro-groups. A lot of them contain anilines, which may have been removed in other screening projects. There are also a range of other functional groups. Chris carried out a scaffold analysis of the fragments. The most common scaffolds are all aromatic systems. Is this because we are using fragments, or is something more important going on? The most common scaffolds in drugs from DrugBank include many of the ring systems identified in the fragments. The PDB presents a similar situation. The same substructures are appearing. BindingDB also shows that aromatic systems and hetero-aromatic systems are very common. We can therefore conclude that aromatic systems are important.

Other common characteristics included:

• LogP and LogD are generally below 3
• Ionised groups are quite common
• There is a large proportion of ‘disc-like’ shaped molecules
• Molecular weight- The screening collection has been culled at 300. The larger ones usually have an iodine. Amongst the hits that have been published, the peak is around 200
• The vast majority of compounds that went into the screens are not ionised
• A lot of the compounds that were screened were not aromatic, but a lot of the compounds identified as hits were aromatic
• The role of increased 3D shape is unproven.

Do we need more 3D structures in the fragments?

It looks like there are a lot of 3D shapes, but actually there are virtually no 3D shapes going into the screening. There is a large proportion of disc-like shaped molecules.

Other things you can find out from the fragment database

• What are the major targets people are screening against?
• Enzymes are the most common, but there are also challenging targets
• Many are active against multiple targets.

Chris presented an example of indazole, which binds to different targets. He has carried out a study to determine whether it binds in a similar manner to the different targets. Read the slides for more information. He also presented work he had carried out on kinase fragments, on the effect of pKa and target type, on the frequency of different amino acid residues in the binding site and whether acidic ligands bind to basic residues.

In conclusion, Chris found that:

• Fragment screening hits tend to be lower molecular weight, contain aromatic rings and ionizable groups
• Some targets (GPCR, Ion channels, PPI) select for specific physicochemical properties
• Detection technology does not appear to influence properties of hits identified
• Measured affinities of fragment hits are in uM to mM range.