Improving “in silico” drug discovery

Drug discovery is a time consuming business and ah don’t mean those Friday nights searchin’ for Berkeley frat parties. Now some Frenchmen have found a new way to do it that don’t involve no dealers, middle men or rolled up $20 notes.

For decades now, drug companies have tried the suck it ‘n’ see approach to drug discovery. But the alarming number of failures are forcing the industry to adopt a less lethal method.

Enter “in silico” chemistry: the process of discovering drugs by sheer number crunchin’ computer power. The task boils down to finding molecules that bind to particular proteins such as receptors, enzymes and ion channels.

There are two methods. The first involves takin’ a look at the things that already bind to the target and then huntin’ for similar molecules that might also bind. That ain’t much more than a glorified Google search.

The second involves modelling the 3D structure of the target and working out what kinda molecule might fit that space, like a piece in a jigsaw. That’s a much harder burden o’ work; those supercomputers get all hot ‘n’ sweaty just thinkin’ about it. And even then it don’t often work cos nobody knows the 3D structure of most receptors.

But Jean-Philippe “Per” Vert at the Centre for Computational Biology in Fontainebleau , France, and his friend, Laurent, have been playin around with an impressive new approach called in silico chemogenetics. The idea is to mine the entire chemical space — which is essentially the set of all small molecules — for interactions with the biological space, meaning the set of all proteins in a particular class of target such as receptors.

Per Vert has used a well-established machine learning algorithm to carry out this task in a way that gets round some of the known limitations (such as having to narrow the search manually to prevent the computer chokin’ on it). He reckons the technique shows a dramatic improvement in the way it predicts potential binding molecules for an important class of receptors known as G-protein coupled receptors. This already qualifies the technique as a valuable tool for drug discovery, he says modestly.

Let’s hope he’s right.

Ref: arxiv.org/abs/0709.3931: Kernel Methods for In Silico Chemogenomics

Comments are closed.