As you pace your pharmacy waiting for someone behind the glass to prep your prescription, time can drag. But did you ever wonder how long it actually takes to create a new drug? About 12 years and $1 billion are spent getting a medication ready. Artificial intelligence (AI) aims to speed that up. The goal: More medicines for rare diseases. Fewer side effects. Personalized treatments. In less time and at a lower cost. At least 230 companies use AI to help develop new drugs. But for all the hype, AI has yet to bring a drug to market on its own. Will it live up to its promise and save lives? For people like Laura Roix, new treatments can't come soon enough. Running Out of Time Every October, Roix got a cough. And every year, her doctor treated her for pneumonia. But the cough would always return. Worried about her family history of lung disease, she went to a pulmonologist, a doctor specializing in the lungs. A scan showed scars in the bottom of her right lung. There was little to worry about; it was probably asthma, she was told. For the next 6 years, Roix relied on inhalers and lozenges to soothe her dry, hacking cough. But it didn't get better. In fact, she started to feel short of breath. Roix knew something was wrong. So she saw another pulmonologist. Finally, she got the right diagnosis: a life-threatening lung condition called idiopathic pulmonary fibrosis (IPF). Over time, IPF scars and stiffens the lungs, making it hard to breathe. If it becomes severe, a lung transplant may be needed. Over the next few years, Roix took oxygen at night. She went to pulmonary rehab. Still, her cough got so bad that she couldn't even talk on the phone. "It's a vicious cycle," she says. "You can't breathe. You're coughing. You're coughing more. You end up with anxiety. And it's just one big mess." There weren't good options. "At the time that I was diagnosed with idiopathic pulmonary fibrosis, there was nothing out on the market," Roix says. She signed up for a clinical trial for a new drug and joined the lung transplant wait list. Roix even went to Washington, DC, to urge the FDA to approve two IPF drugs: nintedanib and pirfenidone, which are now available. But they weren't her solution. Hacking a Deadly Lung Disease Alex Zhavoronkov, PhD, doesn't know Roix. But the CEO of Insilico Medicine, a biotechnology firm, doesn't want anyone to go through what she has. On the day he turned 42, Zhavoronkov announced a possible treatment for IPF -- developed by Chemistry42, his company's AI platform. Insilico's news is one of the latest announcements in the world of AI and drug discovery. But the biggest challenges to treat people like Roix still lie ahead. Part of AI's promise is to improve treatments. With IPF, Insilico says its new drug needs one-tenth of the dose of a current drug, nintedanib, and works in a different way. The company chose IPF as the disease to try to solve with AI because of fibrosis, the process by which your body makes scars. "Your body really needs fibrosis. Otherwise, your scars do not heal and your wounds do not heal," Zhavoronkov says. But in IPF, fibrosis goes wrong. He sees that as one way that bodies break down over time. IPF becomes more likely later in life but isn't a normal part of aging. Insilico's team developed its new drug in 18 months. First, it used AI to study IPF and found a new way to target it. Then it ran algorithms to find new drugs and tested them in a lab. As expected, it took rounds of experiments, with the algorithm learning from each round. The result: a new drug hopefully headed for clinical trials starting in early 2022. If all goes well, it may be on the market 4 years later. Insilico hasn't published its IPF findings in a scientific journal yet. AI helped design the drug. But clinical trials are often longer and harder. Many drugs seem promising in labs and sometimes in mice but fail when tested in people. As Zhavoronkov puts it, AI is like a Ferrari: "You go from 0 to 100 very quickly. But then you move with the speed of traffic." Even if AI doesn't ace every new drug, it can simplify the process. "At minimum, what you can do is just take some of the tasks that currently rely on human intuition and automate them and make them much more systematic," says Regina Barzilay, PhD, a Massachusetts Institute of Technology professor of AI and health. She was part of a team that used AI to make a new antibiotic. They dubbed it halicin, after HAL, the computer in 2001: A Space Odyssey. "The time is ripe" for this work, the team wrote in the journal Cell. For all of AI's promise, there are also pitfalls. There's a balance to strike between data, human biology, and how far technology can go before a person must step in. Picking a Lock Imagine a lock that you need to make a key for. That's what it's like to hunt for new drugs, says Ola Kalisz, a research engineer with U.K. firm ExScientia. The lock is part of a disease. The key is a treatment. "Being given the lock, you try to reverse-engineer what would be the key that cracks open the lock," she says. To do that, AI algorithms learn through trial and error, pick new data points to learn from, predict what existing drugs would work, or recommend how to make a new one. "You just sit away at the computer, coding away the idea that's in your head," Kalisz says. Virtual experiments test whether it's working. "You want to check, 'Did you meet your objectives?'" Kalisz says. But first, you need lots of data -- and the right kind. Or the path to a new treatment might be a dead end. Data Dilemmas Without enough data to search through and learn from, the possibilities are limited. Sometimes, scientists can't get the data they want because it's not public. Plus, biological data is often "standardized," says Andreas Bender, PhD, of the Centre for Molecular Science Informatics at England's University of Cambridge. That is, it's not diverse enough. There can be misses, too. Especially with complex diseases, MIT's Barzilay says, "we don't understand the biology of the disease enough to collect the right data." In these cases, there isn't just one lock that needs a key, but a nest of them. Scientists in this area often know more chemistry than biology, Bender points out. Ultimately, he says, it's important to distinguish between using AI to create a compound -- and actually discovering a treatment that people can take. When Humans Step In In January 2020, headlines touted the first "AI-discovered" drug to start clinical trials. It's a medication for obsessive-compulsive disorder (OCD) made by ExScientia and Japanese pharma company Sumitomo Dainippon Pharma. (ExScientia declined to provide an update on those trials for this story.) Certainly, humans were involved in creating the OCD and IPF drugs. People are still needed to find and make new drugs. The process blends artificial and human intelligence, just as ExScientia's Centaur Chemist platform is named for a mythical creature that's half-horse, half-human. Insilico's Chemistry42 platform has a name that would make any sci-fi fan smile. In Douglas Adams's The Hitchhiker's Guide to the Galaxy, super intelligent mice create a computer called Deep Thought to come up with the meaning of the universe. Deep Thought's answer: 42. For Zhavoronkov and others in the AI field, the number has become an Easter egg reference to artificial intelligence itself. When can AI create a drug on its own? "The question is at which point it can be done very fast and with no humans," MIT's Barzilay says. "And this I really cannot really predict." For Roix, meds ultimately weren't the answer. She stayed in the clinical trial until she got a transplant of her left lung. Her cough is much better. Just 8 months after the transplant, she was even able to finish a 5K. But an organ transplant is a last resort. Many people never make it off the wait list. In Roix's case, her right lung is still at risk. "It's almost dead," she says. When you're sick, you just want to know what will help you. Whether AI made it is probably not on your mind. "I believe that we need more drugs out there to help stop the spread of fibrosis in the lungs or stop it completely," Roix says. "I think that whatever we can do to get more drugs out there, the better off we are." Source
