Scientists Make Digital Breakthrough In Chemistry That Could Revolutionize The Drug Industry

In June, the U.S. government purchased the vast majority of world’s supply of remdesivir—a FDA-approved antiviral treatment for Covid-19—for July through September. Gilead, the company that makes the compound, recently announced that it would meet international demand by the end of October. Yet all along, digital instructions for whipping up a batch of the nearly 400-atom molecule at the push of a button have been sitting on Github, an online software repository, freely available to anyone with the hardware needed to execute the chemical “program.”

A dozen such chemical computers or “chemputers” sit in the University of Glasgow lab of Lee Cronin, the chemist who designed the bird’s nest of tubing, pumps, and flasks, and wrote the remdesivir code that runs on it. He’s spent years dreaming of a future where researchers can distribute and produce molecules as easily as they email and print PDFs, making not being able to order a drug as archaic as not being able to locate a modern text.

“If we have standard way of discovering molecules, making molecules, and then manufacturing them, suddenly nothing goes out of print,” he says. “It’s like an ebook reader for chemistry.”

Cronin and his colleagues described their machine’s capability to produce multiple molecules last year, and now they’ve taken a second major step toward digitizing chemistry with an accessible way to program with the machine. Their software turns academic papers into chemputer-executable programs that researchers can edit without learning to code, they announced earlier this month in Science. And they’re not alone. The team represents one of dozens of groups spread across academia and industry all racing to bring chemistry into the digital age, a development that could lead to safer drugs, more efficient solar panels, and a disruptive new industry.

A chemical computer or “chemputer” sits in the University of Glasgow lab of Leroy Cronin, the chemist who designed the bird’s nest of tubing, pumps, and flasks, and wrote the remdesivir code that runs on it. He’s spent years dreaming of a future where researchers can distribute and produce molecules as easily as they email and print PDFs.
A chemical computer or “chemputer” sits in the University of Glasgow lab of Leroy Cronin, the chemist who designed the bird’s nest of tubing, pumps, and flasks, and wrote the remdesivir code that runs on it. He’s spent years dreaming of a future where researchers can distribute and produce molecules as easily as they email and print PDFs.
Leroy Cronin,
The Cronin team hopes their work will enable what they describe as “Spotify for chemistry”— an online repository of downloadable recipes for important molecules that they say could help developing countries more easily access medications, enable more efficient international scientific collaboration, and even support the human exploration of space.

“The majority of chemistry hasn’t changed from the way we’ve been doing it for the last 200 years. It’s very manual, artisan driven process,” says Nathan Collins, the chief strategy officer of SRI Biosciences, a division of SRI International, a research company developing another automated chemistry system that’s not involved in the Glasgow research. “There’s billions of dollars of opportunity there.”

At the heart of Cronin’s new work lies what he calls a chemical description language or XDL (the “X” is pronounced “kai” after the first letter in the Greek word for chemistry). XDL is to the “chemputer” as HTML is to a browser—it tells the machine what to do. The group has also created software called SynthReader that scans a chemical recipe in peer-reviewed literature — like the six-step process for cooking up remdemisvir — and uses natural language processing to pick out verbs like “add,” “stir,” or “heat;” modifiers like “dropwise;” and other details like durations and temperatures. The system translates those instructions into XDL, which directs the chemputer to execute mechanical actions with its heaters and test tubes.

One of the framework’s strengths, according to Cronin, is that chemists can edit the chemical protocol in plain English. This feature lets researchers operate the machine with little training, and, crucially, harness their chemistry expertise to spot bugs in the code. Chemputer crashes can be serious affairs. “The human will always need to be there to make sure you don’t have a dumpster on fire,” he says.

The researchers tested the system, and no dumpsters burned. The group reported extracting 12 demonstration recipes from the chemical literature, such as the numbing anesthetic lidocaine, all of which the chemputer carried out at efficiencies similar to those of human chemists.

Robotic transformation of chemistry
Cronin built a company called Chemify to sell the chemisty robots and XDL package, although he’s also posted free instructions online for building and programming the machine. And already the device is making inroads in the chemical world. In May of 2019, the group installed a prototype at the pharmaceutical company GlaxoSmithKline.

“The chemputer as a concept and the work [Cronin]’s done is really quite transformational,” says Kim Branson, the global head of artificial intelligence and machine learning at GSK. The company is exploring various automation technologies to help it make a wide array of chemicals more efficiently, but Cronin’s work in particular, Branson says, may let GSK “teleport expertise” around the company. Once a chemist designs a promising molecular recipe, rather than writing up a report or teaching a colleague, they’ll just press the share button.

Researchers say that while Chemify isn’t the most sophisticated automated chemistry platform, it might be the most accessible. It’s built around the traditional tools of beakers and test tubes and functions in the step-by-step “batch” paradigm that chemists have used for centuries. Cronin also intends it to be universal: compatible with any batch chemistry robot. Researchers with their own machines just need tell the software what parts they have and give it figures like how hot their heater can go.

Other groups are betting on a more dramatic break from chemistry’s roots. At SRI, Collins oversees the development of a platform called AutoSyn, which uses an alternative approach called “flow” chemistry. Rather than mixing up a batch of one substance in one beaker, and then moving it to another flask, in flow chemistry reactions play out continuously. Chemicals stream together in tubing, react there, and get carried off. With more than 3,000 pathways, AutoSyn, which Collins and colleagues described in a publication in June, can recreate almost any kind of liquid based reaction.

Doing chemistry in flow requires specialized hardware and extra effort to translate chemical procedures from their batch descriptions, but that investment buys an “exquisite” control over aspects like heat transfer and mixing, Collins says. If machines like AutoSyn can automatically run hundreds of subtle variations on a published reaction, the detailed datasets they generate could highlight the best way to make a chemical.

The literature may be a good place to start, but many published experiments have flaws. Collins estimates that chemists spend 30% to 70% of their time just working out missing details in known reactions. ”[A reaction] is written up by someone who sits down and bases it on their notes from something they were doing the day before, or maybe something they did six months ago,” he says.

While AutoSyn and the Chemputer are both able to reproduce the majority of published reactions today, the next step will be making the machines reliable and “Apple groovy,” as Cronin puts it. Collins says that AutoSyn used to need an engineer to keep it functioning for more than half of its runs, but now needs fixing less than 10% of the time. Eventually, he hopes, users will troubleshoot the system over the phone.

“This is still a very new science,” he says. “It’s started to explode really in the last 18 months.”

One force driving that explosion has been the Defense Advanced Research Projects Agency (DARPA). It’s wrapping up a four-year program called Make-It, of which both the Chemputer and AutoSyn are alumni. The long-term goal of the program’s manager, Anne Fischer, is to speed up the discovery of useful molecules, which has historically involved a lot of waiting around while chemists laboriously smithed atoms into novel configurations. “The slow step is always making and testing the molecules,” she says.

But now that Make-It has helped produce robotic tools to build molecules like the Chemputer, AutoSyn, and others, she’s directing a new DARPA program, Accelerated Molecular Discovery, that looks to the next stage: developing smarter software to tell the robots what molecules to make, and how to make them.

This is still a very new science. It’s started to explode really in the last 18 months.
Nathan Collins
CHIEF STRATEGY OFFICER OF SRI INTERNATIONAL
″We’re now trying now to harness what we’ve done in Make-It and expand it out so we can teach computers how to discover new molecules,” she says.

The secret to doing so, many believe, is machine learning. And some machines capable of rudimentary chemical learning are well underway. Connor Conley, a chemist at MIT, is a member of a team that last year paired an automated flow chemistry system with an algorithm to direct it. The algorithm trained on databases of hundreds of thousands of reactions and was able to predict recipes for new products. “It tries to understand, based on those patterns, what kind of transformations should work for new molecules it’s never seen before,” Conley said.

He stresses that the system has a long way to go. Its predictions were based on similar molecules and human chemists needed to flesh out details missing from the machine-generated outline. Nevertheless, the work supported the notion that software can come up with useful recipes.

MIT is collaborating with more than a dozen chemical and pharmaceutical companies to advance its molecule-predicting algorithms, and some companies have already put the software to use. Juan Alvarez, the Assistant Vice President of computational and structural chemistry at Merck, says that Conley’s machine learning algorithm is one of a variety of chemistry prediction tools that the company has made available to its internal researchers. “It’s absolutely being deployed to impact our timeline today,” he says.

While each group approaches automation from a different angle, they’re all tackling the same problem. A near infinite diversity of possible molecules exist—some of which are surely life-saving drugs or revolutionary new materials—but precious few human beings have the specialized skillset to analyze, make, and test these compounds.

They aim to keep those rare skills from going to waste. In some ways the work of chemists still resembles the work of scribes, who once painstakingly copied and corrected the writings of others. Researchers like Cronin hope that with the chemical equivalents of the printing press, word processor, and autocorrect in hand, tomorrow’s chemists will spend less time recreating, and more time composing.

“It’s not about replacing chemists,” Fischer says. “It’s about giving chemists the tools to allow them to implement and apply the chemistry and allow them to be creative high-level thinkers.”

– CNBC