Assessing the druggability of protein targets computationally is becoming an increasingly useful technique, especially as the hunt for novel drug targets becomes harder and harder. Historically, there has been two approaches taken. One approach analyses the empirical evidence of known drug/ligand interactions and uses them to predict the druggability of other proteins based on structural and physical similarities. Alternatively, the other approach looks for ligand binding potential based solely on structural data of the protein.
A recent publication by Kozakov, Whitty and Vadja in J. Med. Chem takes a slightly different angle on computational druggability assessment, inspired by pioneering work analysing empirical fragment screening data by Hajduk et al.
Firstly, a method was developed for determining binding hotspots on a protein surface by docking a selection of 16 different ‘fragment’ molecules. The fragments were docked as rigid conformers, and hot spots defined as areas where multiple fragments would dock in the same place forming ‘clusters’. The FTMap algorithm to do this has been made available to use online.
They then attempted to determine whether it was possible to reliably predict druggability of a protein based on the size and position of the hot spot clusters as determined by FTMap. They defined a pocket as requiring a primary cluster, and at least one secondary cluster nearby which could also be involved in the ligand binding interaction. From this, they devised three metrics by which they would measure their pockets by; the strength of the primary interaction (S), the distance between primary and secondary clusters (CD), and the overall size of the resulting ‘pocket’ (MD) (see table below for more in depth definitions).By comparing the FTMap generated cluster locations of known drug targets, they were able to determine values of S, CD and MD that would predict a protein to be druggable (shown in the table above). By requiring that S > 16, CD < 8 Å, and MD > 10 Å, they were able to correctly classify a surprising proportion of the proteins tested known to be druggable, with only a handful of proteins with known inhibitors not satisfying their criteria at the drug binding site.
This technique seemed to particularly shine when it came to testing the druggability of more non-conventional drug targets. The authors defined a set of slightly relaxed rules which allowed the prediction of protein druggability using non-canonical drug types, such as macrocycles or polypeptides. This allowed the technique to be used to predict the druggability of known macrocycle binding proteins and protein-protein interactions, with a fair degree of accuracy. Other computational techniques may have less success in this area due to the large size and shallow nature of the pockets in these classes of protein target.
Overall, it seems that by using computationally determined hot spots as a measure, the authors have developed an effective and novel approach for assessing the druggability of proteins. The breadth of applicability of this technique compared with others should hopefully be of particular help with the increasing trend in drug discovery of looking towards more unconventional drug targets such as protein-protein interactions.
Blog written by Tom Moore