ChEMBL

This is now almost a classic case of drug design, so much work has been done in the area, and frankly quite abysmal translation through to effective drugs. Some of the preclinical work I was involved in is here (subscription required). Thrombin was a tough target to get right, both from a chemistry, and as it also turns out, a biology perspective. Back to the PPI subject though - there are two high efficiency pockets in thrombin - the 'P1' pocket that binds the Arg/Lys of substrates, and an 'aryl site' that binds hydrophobic (and generally as the name would suggest aryl groups). In between these two pockets is a 'saddle' of thrombin main-chain that needs to be traversed (the Donor - Acceptor - Donor annotation) - and for our purposes there is this interesting donor acceptor donor feature complementary to main chain hydrogen bonding features (which are used to bind the peptide main chain in substrates). It is notable that these DAD and ADA (and usually not too much longer) triplets are really common in PPIs, and have a certain recurrent spacing between them - complementary to the spacing of the restricted composition protein receptors for most drugs. In my view, there will be 'magic' - or more accurately useful spacings of donor-acceptor features in peptidomimetics - and more importantly a barren far larger wasteland of chemical sensible and possible donor-acceptor spacings, that just don't fit the majority of protein receptor possible interactions - in fact it's interesting to speculate whether this acceptor-donor spacing is a useful parameter for chemical space display. You can imagine an interesting thing to look at would be to take a set of pharmacophores and then scramble feature properties and investigate whether close D-D are rarer than they could be, D-A commoner, etc., ....

I said the thrombin example is a little bit like a saddle - well here's a view looking down the backbone ridge of thrombin which hopefully makes in clearer - we can only see one of the donor features in this view, other is occluded due to the curve shape of the backbone.

Of course, this particular example is a beta-mimic, and there are many other classes of PPIs - some involving the main chain (primarily beta-strand and turns as we'll see in a later post) and others not involving backbone interactions (primarily alpha-helix mimics) - Remember though that these hydrogen bonds don't buy you much potency under usual conditions, they are almost a necessary evil with respect to potency, so we don't want to got too mad in factoring them into our designs as a source of affinity - but they do help lower logP which is always a useful thing - so small polar scaffolds are the best thing to aim for. Loads of people will be incensed by the previous simplifications and assumptions - well this is a blog post, not a paper, so feel free to add comments if you fancy!

One site of particular interest in this area is the TIMBAL database from the group of Tom Blundell.

Atomic interactions and profile of small molecules disrupting protein-protein interfaces: the TIMBAL database - these contain a bunch of specific examples that we wish to mimic with this PPI library.

%T Atomic interactions and profile of small molecules disrupting protein-protein interfaces: the TIMBAL database
%A A.P. Higueruelo
%A A. Schreyer
%A G.R. Bickerton
%A W.R. Pitt
%A C.R. Groom
%A T.L. Blundell
%J Chem Biol Drug Des.
%D 2009
%V 74
%P 457-67

Oh, I think I managed to convince my train-buddy Roman that this is an interesting problem, and he is planning to look into another part of the analysis - which is good news.