This article is part of Fast Company’s editorial series The New Rules of AI. More than 60 years into the era of artificial intelligence, the world’s largest technology companies are just beginning to crack open what’s possible with AI—and grapple with how it might change our future. Click here to read all the stories in the series.
Last December, a conference of biologists gathered in Cancun, Mexico, to review a shocking finding. DeepMind, Alphabet’s artificial intelligence lab and sister company to Google, had beat a roomful of biologists in a contest to predict the shape of a protein based on its genetic code.
That might not sound monumental, but understanding the way proteins fold into three-dimensional shapes is crucial to helping create drugs, which often fight disease by latching onto proteins and altering the way they work in the body. DeepMind was able to predict these proteins’ shapes with significantly more accuracy than the many esteemed academics and professionals at the conference.
“It dawned on me that this is a field that people have been working in for decades,” Mohammed AlQuraishi, a biologist and researcher at Harvard who participated in the contest, told Vox. “The fact that a new group could come in and do so well, so quickly—I felt bad because it demonstrated the structural inefficiency of academia.”
It was startling moment for the drug discovery business: Could an outsider with little experience in biology really barge in and do science better than the experts?
DeepMind’s discovery has called into question whether major pharmaceutical companies can maintain dominance in their own industry if they have to go up against Alphabet, which has steadily built up its credibility as an artificial intelligence powerhouse.
But this isn’t just about a younger, more nimble giant unseating older, lumbering giants. The reality is that a host of new companies are racing to change the way drugs are made, using AI as an accelerant for research. With it, researchers might narrow in on life-saving molecules in a matter of weeks rather than months or even years.
Today’s drug behemoths are aware of the promise more than anyone. They’re investing in artificial intelligence labs of their own and, along with venture capitalists, are pouring money into drug discovery startups with an AI bent. According to Pitchbook data, U.S. drug discovery companies raised $9.4 billion in 2018, and they’ve raised another $4.4 billion so far this year.
Regardless of who wins the drug discovery race, artificial intelligence will impact how drugs are made for good.
Big Pharma’s strategy to maintain dominance
Big Pharma started using artificial intelligence long before DeepMind’s winning moment. But faced with increased competition from DeepMind and a host of upstarts, many pharmaceutical companies are investing externally in startups—even if it means possibly funding companies that they will ultimately compete against.
For instance, Johnson & Johnson was among the investors for BlackThorn, which focuses on developing drugs for psychiatric disorders, when it raised $75 million earlier this year. The company uses brain images to better understand how potential drugs will impact mental state. BlackThorn will begin Phase II clinical trials on a potential drug to treat major depressive disorders at the end of the year that, if ultimately approved, will directly compete with a drug that Johnson & Johnson’s pharmaceutical arm recently launched.
Johnson & Johnson has also signed deals with another company called BenevolentAI, which largely uses scientific literature to train its algorithms to find the right targets in the body. The dueling investments reveal that Johnson & Johnson is placing bets widely and defensively. And it’s not alone: Pharmaceutical companies like Novartis, AstraZeneca, and GlaxoSmithKline, among others, have all signed arrangements with artificial intelligence upstarts. Should any of these companies come up with groundbreaking formulations, legacy pharmaceutical companies are already poised to benefit. They might also be positioned to acquire a company they’ve already worked with or invested in.
After all, while startups have the technology, Big Pharma has the size and the money to bring many of these experiments to life. In this way, they can work together to beat out potential threats like DeepMind.
Rebuilding pharma from the ground up
Startups in this field have their own prerogatives. Some young companies hope to become the new powerhouses for research and development, leaving clinical testing to the big companies. Others are currently partnering with Big Pharma, but hope that they will eventually be able to displace the behemoths of drug discovery entirely.
One newbie drug hunter is Insilico, a biotechnology company with a focus on longevity. In September, it revealed that it had used artificial intelligence to design a potential drug in 21 days. The process only cost $150,000, a small sum in the world of drug discovery, and took days instead of years—exemplifying the promise of using machine learning algorithms to create new drugs. It has so far shown promise in mice. Subsequently, the company raised $37 million and plans to partner with pharmaceutical companies to run tests in humans.
Insilico trains its machine learning systems on existing research and scientific studies, but relies on partners to run clinical studies. This strategy serves to augment the existing pharmaceutical industry with tools that can surface potential molecules faster. Then, it relies on those companies to then validate that potential drugs actually work.
Some researchers are debating how groundbreaking Insilico’s work really is. “The reality is . . . it’s not a leap,” says Chris Gibson, CEO of Recursion, an AI drug discovery startup (and one of Insilico’s direct competitors). He says that while Insilico’s discovery is big—a new and novel molecule that wasn’t previously known—it’s still an incremental improvement because the new molecule is only slightly different from inhibitors scientists already knew about. Other critics have pointed out that human researchers might have been able to find this molecule in a similar time frame.
Gibson sees issues with DeepMind’s protein-folding prediction as well—one that has been echoed by other researchers in the field. He says that while it is impressive, it only contributes a piece of the overall drug discovery equation. Gibson is skeptical about how much can be accomplished with artificial intelligence alone, even though he considers himself bullish on the technology.
He thinks to make real progress in drug discovery, you need to remake pharmaceutical companies from the ground up. Rather than pharmaceutical giants outsourcing research and development to an AI startup, scientists should work in tandem with artificial intelligence as they seek out new drugs.
Part of the reason he believes in this approach is because he thinks to get the best results you need to generate data that is tailor-made for artificial intelligence to digest. He argues that drug discovery companies using existing information to train their algorithms may find their results fall short in the long term because the data they’re using was written without machine learning in mind.
Even with the promise of artificial intelligence, creating a new drug remains a daunting proposition.
“The real challenge for our industry, for over a decade, is that 90% of the drugs we put in the clinic fail, and that means we’re wrong 9 out of 10 times with a target and a chemical,” says Gibson.
In order for AI to prove its worth as a tool, it can’t just meet that same 10% success rate—it has to best it. Artificial intelligence has already changed the drug-making process, but if DeepMind ever does replace biologists, it won’t be anytime soon.