Molecular recognition and chemical space navigation in drug discovery

Abstract

Efficient discovery of bioactive molecules is an essential goal of Computer-Aided Drug Design (CADD). The molecules can be used as chemical probes, to validate novel targets, or as starting points for drug discovery. This endeavour is particularly challenging in the case of proteins that are considered undruggable or for which no ligands are known. These are precisely the type of proteins that must be targeted in order to expand the “druggable genome” and extend the range of therapeutic opportunities.

CADD tools available nowadays are numerous but have limitations that must be overcome in order to improve the efficacy and efficiency of drug discovery. Particularly because they should also be able to exploit non-standard sites, such as protein-protein interfaces, allosteric sites or cryptic pockets. They should also be adapted to address specific needs in the drug discovery process. Finally, they can be used to gain a fundamental understanding of the behaviour of molecular systems and the rules of molecular recognition that govern the recognition of a drug by its target. In this thesis, I have explored each one of these aspects.

Initially, I developed an automatic pipeline that can be used in Fragment-Based Drug Discovery (FBDD) to navigate the “fragment chemical space”. Starting from a fragment hit with a known binding mode to its target, the platform automatically seeks non-obvious analogues (scaffold hops) within large chemical collections, delivering fragment hits, with novel structures that would, otherwise, be missed. I validated the platform using a fragment hit of the first bromodomain of the Bromodomain-containing protein 4 (BRD4) taken from the literature as a starting point. The platform identified multiple fragments with novel scaffolds and excellent ligand efficiencies. For some, their binding modes could be corroborated experimentally.

The optimized fragment identified in the first study allowed us to investigate the unusual behaviour of structural water molecules in BRD4(1) and their role in molecular recognition. Paradoxically, a hydrophobic binding hot spot of BRD4(1) is lined with water molecules. A series of compounds were derived to probe the preference of this site for chemical groups with various degrees of polarity. Molecular dynamics (MD) and free energy calculations allowed us to rationalize the experimental results.

I have then used de novo design (DND) methods to further grow the most active fragment into a very potent and efficient drug-like BRD4 ligand.

Finally, I have discovered the first ever described inhibitors of the Three Prime Repair Exonuclease 2 (TREX2) protein.