25 July 2012
Linking computers together to form a vast global network transformed computing – and then the world. Now, by cramming nearly 250 years’ worth of chemical knowledge into a single software package, a team led by Bartosz Grzybowski has created a chemical internet called Chematica that could do the same for substances, from drugs to fuels to plastics.
Rather than individual computers wired together physically, the nodes are 7 million different substances, linked by knowledge – the reactions that get you from one to another. “Think about it as a collective chemical brain,” says Grzybowski, who is a chemical engineer at Northwestern University in Evanston, Illinois.
This chemical rewiring is already revealing cheaper or more efficient ways to make existing drugs, as well as the ability to spot easy ways to make potentially dangerous chemicals before criminals do. If taught to learn new rules, as well as to restructure existing knowledge, the software might even throw up ways to make totally new compounds with useful applications.
So what novel things does the chemical internet enable? To some extent chemistry is like cooking – mix a few ingredients together and out pops something new. Just as there is more than one way to rustle up a meal, chemists have a staggering number of options available.
Most routes to obtaining a particular molecule or drug involve breaking down this target into simpler molecules and then figuring out how to make those. As there are multiple ways of executing each stage, the number of possible recipes quickly balloons and chemicals end up being produced via a patchwork of tried-and-tested methods, which might not in fact be the optimal route.
“People use well-trodden paths, they have to,” says Lee Cronin, a chemist at the University of Glasgow, UK. “No human can ever approach considering which route is better, it’s absolutely ludicrous for a human to even think this way,” Grzybowski adds.
It’s easy, though, for his team’s software. Chematica provides a map that algorithms can scour to find more optimal routes. Building this vast interconnected database took Grzybowski and his team about a decade, reflecting the vast quantity of information they had to gather. “We compiled information from different sources, journals, databases,” he says. “The first five years were just compiling the database.”
The team also programmed Chematica with over 86,000 rules – such as the need for or absence of water, or the required balance between acids and bases, the compatibility of various solvents or whether certain substances create oxidising or reducing conditions. That took a further two years: Grzybowski likens it to translating the brains of five people into big tables of numbers.
The work may pay off quickly, though. These capabilities have already proved useful in hunting down “one-pot reactions”. These much-sought reactions are the chemists’ equivalent of the slow cooker: add all the ingredients at the start and create a delicious new substance without having to go through intermediate stages in which compounds must be isolated and purified. “In chemical production worldwide, 80 per cent of the cost is purifying intermediates,” says Grzybowski.
He found that Chematica was able to flag up over 1 million previously unknown one-pot reactions – in particular for a family of compounds that are targets for asthma treatments. Previously, these could only be made in reactions with multiple, separate steps each requiring laborious and costly purification of the product before the next step could begin.
When Grzybowski’s team tried out the most striking of these one-pot reactions – previously a four-step process – they found that nearly half of the raw material was converted into the final product. That’s roughly double the yield that is obtained the four-step way (Angewandte Chemie, DOI: 10.1002/anie.201202155).
“It’s a very elegant result, it shows there is potential to be unlocked,” says Cronin, though he would like to know if the algorithm can scale to reactions with hundreds of steps.
Chematica could be useful for finding more than one-pot reactions. It can work out cheaper ways to make the same substance, depending on conditions, such as labour or raw material costs, which can vary with location.
The ability of the software to explore multiple pathways also lets it take steps that would be unintuitive to a human chemist, such as starting with a more expensive material that opens up access to much cheaper ones.
When Grzybowski tried the software on 51 products manufactured by ProChimia Surfaces, the Polish chemical company that he owns, it came up with alternative syntheses that would drop production costs from an average $39.60 per gram to $21.50 per gram, a saving of about 45 per cent (Angewandte Chemie, DOI: 10.1002/ange.201202209). The software can also be optimised to avoid other factors such as environmentally harmful chemicals.
Grzybowski says he has already sold Chematica to one pharmaceutical company and is in talks with two others. “It is unlikely to replace the human role in deciding a route, but it is clearly able to consider many more possibilities than a single person or even an entire company could,” says Robert Paton, a chemist at the University of Oxford.
Realising these theoretical savings might be hard though, says Cronin, as switching chemical pathways isn’t that simple. “The chemical industry is going to be slow to respond, because if you’ve already invested $30 million in a plant, you’re not going to tear it down.”
Still, synthesis isn’t the only potential application of the software. Grzybowski is exploring how Chematica might be used to spot potential terrorists building chemical weapons from seemingly innocuous substances.
He suggests using Chematica in situations where authorities are already monitoring suspected criminals, to flag up dangerous combinations of chemicals in their possession. These might be so ordinary that a human would not spot their potential.
To demonstrate, Grzybowski set his team the task of finding a way to synthesise lethal VX nerve gas from substances available in a supermarket. Chematica revealed that nine ordinary ingredients, including water, salt and ammonia, form a relatively short chemical path to creating this gas (Angewandte Chemie, DOI: 10.1002/anie.201202210). “I don’t know whether [Osama] bin Laden was aware of this, but I didn’t know,” he says. Grzybowski adds that he is currently working with government agencies to harness Chematica’s security potential.
There’s also the possibility that a terrorist could use Chematica to help them build a chemical weapon, though Cronin argues that it’s always better to have the knowledge in the open. “If anyone wants to make a chemical weapon right now, it’s easier than ever to do so,” he says. “This approach allows us to keep an eye on that in a very open and intelligent way.”
Perhaps the most exciting application for Chematica lies in using it to come up with new compounds, something Grzybowski is still working on. “We’re improving the chemical rules, we’re implementing machine learning,” he says. “If we play it right, it’s not only going to be learning from the past but also projecting into the future, telling people what they should be making to make the entire discipline most efficient.”
Could Chematica really innovate in this way – reducing chemists to mere fast-food chefs? Paton is sceptical: “It would struggle to find a genuinely novel molecule,” he says. “Organic synthesis is considered an art and a science, and there is a lot of scope for creativity when people design new reactions.”
Grzybowski thinks that just means he has work to do in convincing people of his software’s potential: “I expect tremendous opposition to these ideas. I’m ready for the battle.”