While drug companies struggle to develop medicines for rich countries and typically overlook diseases elsewhere, a robot scientist named Eve has found compounds that could fight drug-resistant malaria. Eve’s developers believe their artificial intelligence (AI) technology could speed up drug discovery, as critics call for a “match” with a live chemist.
The AI-endowed robot is designed to add a new, advanced ability to learn on top of the computational smarts that the pharmaceutical industry already uses. In research published February 4 in Journal of the Royal Society Interface computer scientist Ross King from the University of Manchester in England and his team say Eve found a chemical called TNP-470 effectively targets an enzyme that is key to the growth of Plasmodium vivax, one of the parasites that causes malaria. “I didn’t expect to actually find any useful compounds, I thought we’d just demonstrate the AI,” King says.
Eve not only has brains—it also has drug discovery brawn. Its computer server controls two robot arms that dance amidst equipment for dispensing liquids into plastic plates containing large numbers of wells. The plates are used in screening tests for potentially useful drug compounds. Drug molecules, in essence, act like tiny keys that slot into protein or enzyme locks. In the plates each well holds a biochemical system containing a lock, and when a key fits into it the system triggers a detectable signal such as fluorescence. Initial signals—or hits—are usually found on one instrument before further tests are done elsewhere to check if the key really does fit the lock. Eve integrates these usually separate capabilities, accelerating the research process.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
Pharmaceutical companies often have to screen hundreds of thousands of compounds to find hits that tell them about the nature of the lock. These hits are never the exact, perfect key needed to treat a disease. So, after having slogged through the lengthy screening process chemists and biologists spend considerable effort using the data to work out what compound to make and test next. To do so, they create a “quantitative structure–activity relationship” (QSAR) from the screening results. This is a mathematical function that relates a molecule’s composition, shape and properties such as electrical charge to how well it fits the lock. Even with computer-assisted QSAR, however, it still takes a slow trial and error process to hammer out the exact key that could become a drug.
Eve’s artificial intelligence—a set of “active learning” algorithms added onto QSAR capabilities—is set up to find promising leads faster. After working through a “learning set” of around 5,000 molecules the AI gleans characteristics of keys that fit best with the locks. Then it uses those characteristics to predict which remaining members of the collection are more likely to be hits, selecting and testing only them. Since screening the rest of the collection will not add much more knowledge, Eve can quit early.
In the search for a malaria drug one possible lock scientists have targeted is an enzyme called dihydrofolate reductase (DHFR), which plays an important role in cell growth. It is found in people as well as parasites, however, so fighting malaria without harming patients means hitting the parasite version while avoiding the human one. Eve rapidly screened molecules using brewer’s yeast cells genetically engineered to contain DHFR-encoding genes from P. vivax—as well as from P. Vivax species that resist an existing malaria drug—and other yeast holding human DHFR genes. It turned out that among all molecules tested, tiny amounts of TNP-470 could stop yeast with drug-resistant P. vivax DHFR from growing whereas it took concentrations 1,000 times greater to stop yeast with human DHFR.
Eve’s ability screen quickly also reduces costs, because commercial screening compounds are sold for around $15 per milligram, over 10 times the price of gold. “If you screen the whole library, you’ll find all the hits, but you’ve consumed some of all the compounds and a lot of time,” King says. Doing things Eve’s way could help solve what he calls pharma’s “fundamental problem”: It is too slow and expensive to develop new drugs.
Derek Lowe, a medicinal chemist who writes a blog about drug discovery called In the Pipeline, agrees that the potential speed and cost advantages are important, but with a caveat. Eve “seems like a natural consequence of the technologies we use already,” he says. “As with any automated system, my worry about Eve is that people will take its results at face value, never bothering to look under the hood,” he adds. “So while I'm not hostile to this approach, I am wary of it, because I think that many real-world applications—at least the way the real world is now—will be a rough fit for it. But that situation is bound to change over time, and if people aren't working on this sort of thing, it'll never improve,” he says.
Sean Ekins, a computational drug discovery consultant at Collaborations in Chemistry in North Carolina, is much more skeptical. He notes pharma companies have found hundreds of antimalaria compounds more potent than TNP-470 and says that he is not convinced Eve can do QSAR. He wants to see Eve go head-to-head with a real computational chemist. “Eve should go back to the Garden of Eden and leave drug discovery to scientists who know what they are doing,” Ekins says.
King also wants to see a direct competition but says that it is hard to organize such a “bake-off.” He’s used to flak from pharma researchers but still feels it is inevitable that systems like Eve will play a bigger role in finding drugs. “In the future AI systems will be deciding what to make and screen because they’re just better at some tasks than humans,” he says.