We need a better way to discover blockbuster treatments: it seems the more money companies spend on research and development and the more we understand about the genome, microbiome and other "omes", the harder it becomes to bring new drugs to market. With the cost of delivering a novel drug reaching a staggering $2.6 billion (£2.1bn), according to the Tufts Center for the Study of Drug Development, and gestation times of a decade or longer, many are worried about the future of pharma. Now, however, we are on the brink of harnessing the power of computers to customise the right drug to suit the right patient.
It realises a dream first articulated by Britain's greatest theoretician, Paul Dirac. In 1929 he said that, because of advances in quantum mechanics, "The underlying physical laws necessary for the mathematical theory of a large part of physics and the whole of chemistry are thus completely known." So why not use computers to put these laws to work to design drugs? After all, we know aspirin works because it inhibits COX enzymes, rather like a key slides into a lock. Why not try to model how various potential drugs interact with targets in the body?
One problem: quantum theory works beautifully with small numbers of particles, but it's hard to use in real-world examples, where the numbers of atoms or molecules often exceed a septillion (that is one followed by 24 zeroes). This is the "multiscale problem" – for example, we still can't work out how the boiling point of liquid water emerges from its molecular properties.
The multiscale problem rules time as much as space: computational chemists are able to chart what happens to all these molecules down to the level of less than a quadrillionth of a second (a femtosecond). But you'd have to perform a billion or so to get into the microsecond regime, which is the timescale where things become interesting in most real-world chemistry.
Many expected the deluge of information from genomics would change everything, but there is currently not enough deep understanding to efficiently exploit all this data.
Another problem is that in reality, the "keys" and "locks" are complex 3D molecules that vibrate, rotate and deform. The tiniest change in the rotation, shape or angle of attack of a drug molecule can have vastly different effects on its ability to bind with a target. Classically, one explains this with chaos theory, which says that a tiny cause can magnify over time to produce huge effects.
Even with a lot of data about how drugs work, chaos means the way a drug molecule docks with a target site in the body is extraordinarily sensitive to many factors: that is why, using traditional computer methods, independent simulations of the action of an HIV drug on the same target molecule produced different results.
Despite these challenges, in silico drug development is now making important progress, according to Peter Coveney of University College London (UCL). He leads the international CompBioMed consortium that carries out simulations in medicine using supercomputers such as ARCHER, the UK national supercomputer, Prometheus in Krakow, Poland and SuperMUC at the Leibniz Supercomputing Centre, Germany.
One study by Coveney's team, published with GlaxoSmithKline in the Journal of Chemical Theory and Computation, tested a method developed by UCL to rank how well a range of widely varying drug candidates stick to a section of a protein (a "domain") known as the bromodomain, which is linked to cancer and inflammatory disease. Gratifyingly, the results were rapid and reproducible.
Once the team has found a promising drug candidate, they can put it to more stringent testing with a new approach called "Monte Carlo alchemy", a reference to thermodynamic integration with enhanced sampling or, informally, a reference to Monte Carlo methods, which are used in weather forecasting to explore random variations of starting conditions to figure out which parts of the forecast are reliable.
The researchers focused on transformations from a well-understood drug molecule, via unreal (alchemical) intermediates, into an untried molecule to calculate how well the latter works as a drug. They reported they can accurately use Monte Carlo alchemy to give insights into how common mutations cause acquired resistance to anti-breast-cancer drugs such as Tamoxifen and Raloxifene. "Just as in modern weather forecasts, we ran ensembles of simulations," Coveney explains. "And like a forecast of weather, they provided a forecast of how well a novel molecule will bind with a site in the body."
Your doctor will eventually be able to design precisely the right drug to suit you by simulating the way it works using number-crunching services online. One day, not just your head but your whole body will be in the clouds.
This article was originally published by WIRED UK
