The highly requested and long-awaited interaction search is finally here! Given the great amount of structural data we have at hand, the ability to mine ligand-protein complexes based on the interactions they create was a missing piece in our analytics toolbox. The scientific applications of this new feature are numerous:

• Idea generation: finding new chemical matter that fits your expected interaction pattern.

• Scaffold-hopping: finding scaffolds that fit the interaction pattern. 

• Off-target Profiling: finding putative off-targets (binding sites with same interaction pattern)

While implementing the interaction-based search, we encountered several technical and scientific challenges that we had to overcome. I thought I would share some of them here. 

To start with, ligand-protein interactions need to be well characterized. We wanted to go beyond the classical H-bond, hydrophobic, aromatic, and ionic interactions so we implemented our detection method for protein-ligand interactions. This makes things fuzzier and more flexible than standard methods. Our approach allows us to capture halogen bonds, cation-pi interactions, T-shaped interactions with aromatics, etc. On top of this, we also register protein-ligand contacts that don’t fall into any of the standard interaction categories. Our aim with this is to be able to use the data in future applications, e.g. machine learning approaches based on the complete ligand environment.

The data storage presented a second technical hurdle. The information needed to be structured in an efficient way to allow fast querying.

Another challenge came from the fact that the search queries need to respect a set of geometrical constraints. We consider the interactions as vectors and they should be similarly arranged in space between the query and the hits.

Lastly, the user-interface needed to allow very flexible searches. The user should be able to modify the initial search query, e.g. by replacing a hydrogen bond donor by an acceptor type of interaction.

Now, these hurdles have been cleared and we are finalizing the new release. 

Let’s take a look at the interaction search in action. This is what you will be able to do:

• Select several interactions in the 3D Viewer and prepare a query based on these.

• Finetune your query by modifying distances, angles, interaction types, and interactant types (i.e. the atom type of the receptor or the ligand).

• Once results are back, superpose the hits to the query in a single click. The overlay is based on the selected interactions and allows you to very easily compare the ligands and decide what to design next.

Check out the short video in this article showcasing the interaction search and stay tuned for the next exciting article that will come up! 

3decision’s blog has been moved!