Derek Lowe. Credit: T. Hashemi

Computational drug design has always seemed like a good idea. Compared to traditional drug discovery, it seems to offer a way to soar over what can be a trackless swamp. Finding a lead structure from millions of possible drug molecules is no small undertaking in itself. And once found, turning such a lead into a plausible drug candidate is notoriously difficult. That level of success, though, merely buys you the chance to spend your real money in the clinic, where 90% of the compounds that go in never come out.

Far better to sit at a keyboard in some quiet room, hit the 'run' button, and come back later for the answers. The greatest period of enthusiasm for this vision was probably during the late 1980s, when advances in both hardware and software put molecular-modelling technology into more hands than ever before. But the new converts soon found themselves up against a whole suite of challenging problems. The shapes of drug molecules, how they change in both the presence of water molecules and their protein targets, and the shapes of the proteins themselves all had to be dealt with.

The act of 'docking' a drug candidate computationally into its target protein has to take all these factors into account, and fundamental problems remain. Even so basic (and crucial) a thing as the weak hydrogen bonds between atoms can be very difficult to model realistically. Then there's the dynamics-versus-statics problem: drug molecules and their binding targets never stop moving, folding and flexing. Modelling this realistically is hard, and increases the computational burden substantially.

Billionaire investors

Schrödinger, headquartered in New York and a respected company in the field, occupies an unusual niche. A number of other companies, such as their biggest competitor, Accelrys, based in San Diego, California, exist in the computational chemistry area. Unlike Schrödinger, however, very few have attempted to produce a broad range of computational products that attempt to address exactly the sort of problems listed above. Many other modelling-software vendors have left the field over the years, apparently because they could find no good way to make money at it.

The Gates investment fits in well, making Schrödinger one of the few companies independently funded by two billionaires. ,

Schrödinger, however, remains a private company, and in some ways is the most academic of the drug-discovery software vendors. It has had the benefit of a long relationship with David Shaw, a Wall Street investment manager with a scientific modelling background who has kept a hand in the area. In that sense, the Gates investment fits in well, making Schrödinger one of the few companies independently funded by two billionaires. The company does not release financial data, but has said that its revenues are somewhere above US$20 million per year. In that environment, a $10-million investment is substantial.

The company is likely to use its extra dollars for long-term projects: given the complexities of molecular modelling, all meaningful improvements will probably fall into that category. One possible project would be to collaborate with industrial drug-discovery organizations to give the company's products continual real-world testing, with the results fed back to the programmers for further rounds of improvement.

Patience pays

For now, the dream of sitting down far away from all the flasks and the rats and coming up with a viable drug candidate — let alone a viable drug — remains just that.

Molecular modelling has helped many drug-discovery programmes, and probably hindered others if the truth be told, but it has yet to cause drugs to appear where there were none before. And not all of the drug industry's biggest problems are amenable to computational solution. So if Gates (or those following his moves) are expecting dramatic changes, they're likely to be disappointed. But the things that can be computed are still important, and they could certainly be done much better than they are now.

Is enthusiasm for computer-driven drug design on its way back to the peak of the late 1980s? Probably not — but that's partly because people are in no mood to repeat that cycle of hope and disappointment. The Gates announcement more probably represents a realization that if computational drug discovery is ever going to reach the heights expected of it, then the journey is going to be a long one. And it will require money from people who are not expecting and do not need an immediate return.