 we can be slightly more targeted with Qsar. So assuming that I have a binding site where I have at least one molecule that binds, and again, I have a crystal structure of this binding site. Maybe this is a molecule that binds and I have that hydrogen bond partners up here and a certain number of aromatic rings. If I just use some small molecules here to identify the properties, I should be able to describe this molecule. But rather instead of describing this with all the bonds and bond angles and everything, I could just describe this schematically. So let's say that the blue parts here are, for instance, aromatic rings and the yellow ring. No, these would be the rings. And then I might have hydrogen bond donors and the hydrogen bond acceptor or something. In principle, I could just say that, you know what? This is a five-member polyhedron. And then I say what should the distances between all these parts be? And then I could search a database that is just describing a total of molecules in terms of distances. This concept, this simplified model, is called a pharmacophore. So again, you can think of this as a profile of the culprit that we think did it. Because again, I'm not interested in the dual atoms. The question is, can I take this molecule that did have some affinity, distill this down to the very core? What are the most important properties for this to bind? And then I test more molecules and more molecules and then I gradually refine this to find what are the three or four things most important in my molecule to get this to be an efficient binder? And then hope that I can either find more of that in databases or sit down in front of the computer and design a molecule that fulfills this even better. We've done that for a number of small compounds. Let me show you.