The maniacally focused founder of Insilico Medicine is going all in on AI to discover new medicines and extend lifespan.
The path from pandemic to vaccine in only a year was a pharmaceutical feat one is tempted to take for granted. But the truth remains that a typical drug requires more than a decade to design, test, and approve. Even if you know what molecule to target to treat a disease, it often takes several years just to optimize a drug’s chemistry before testing it in clinical trials.
But researchers are now using artificial intelligence to accelerate parts of that process by orders of magnitude. The same kind of technology that recognizes your face or autocompletes your emails can potentially add years to your life, experts say.
One company that has embraced AI for drug design perhaps more than any other is Hong Kong-based Insilico Medicine, a startup that has raised more than $400 million in funding since its founding in 2014. They’ve published more than 130 peer-reviewed papers and built a suite of tools that they and other companies use to search for treatments. They’re currently conducting a phase 1 clinical trial for a small molecule they designed to treat the lung-scarring disease idiopathic pulmonary fibrosis, which they developed at a COVID-19-vaccine-like pace. It took them just 30 months to identify the target, design a drug candidate, and conduct preclinical testing before moving into human trials.
Insilico was founded by a scientist named Alex Zhavoronkov, whose mission is not just to cure individual diseases but to increase longevity. In October, he announced a longevity pledge to dedicate himself, body and wealth, to this end. He’ll give “everything I have now, and what I will get in the future, to only one cause—extending healthy productive longevity for all human beings,” he wrote in an announcement.
Insilico is only one tool to achieving that goal. He’s creating a whole ecosystem around the cause. Kristen Fortney, the CEO of the biotech firm BioAge and a consultant for Insilico, says one accomplishment that has impressed her the most is Zhavoronkov’s creation of the Aging Research and Drug Discovery (ARDD) conference, which will convene for the tenth time next year. “He had a vision of bringing leading academic and industry voices together in a meeting that would chart the future of the field, and he kept at it until he’d achieved it,” she says. “And that’s Alex in a nutshell. He will start with what sounds like a wild idea and then get it done.”
Lives cut short
Zhavoronkov has wanted to increase human lifespans since he was a child. He was born in Riga, Latvia, at the time part of the Soviet Union, in 1979, to a father who worked in construction and a mother who was an engineering architect. His younger sister is now in finance. “Unfortunately, she decided to take a more traditional route,” he says. “But maybe fortunately, somebody in that family has to be normal.”
“I decided to focus on science rather than being religious.”
“When I was a kid, I was pretty ambitious,” he says. He tried to do well in school. “I never expected to be where I am today, but I always wanted to go after aging. Because that’s the biggest problem that I saw, even at that time.” Early on, he was deeply Christian, but religion couldn’t appease a biting sense of injustice. “We’re given a life, we grow, compete, develop, reproduce, and then lose everything we make and die,” he says. “That period of losing disturbs me greatly. And my question was: Why? So I decided to focus on science rather than being religious.”
Losing religion in turn motivated him to find cures for aging. “If we don’t have religion to comfort us, we need to think about other means,” he says. He explored everything, even alternative medicines. “In the 90s, when I was growing up, there was a lot of snake oil. Pretty much like today, but more.” He looked at diets, exercise, and sleep modifications and saw that nothing worked. In high school, he says, “I looked at a lot of BS, and I thought that computer science would be a really good area to structure knowledge and help understand aging.”
Zhavoronkov moved to Canada and earned two degrees at Queen’s University in Kingston, Ontario. He majored in computer science and management of information systems. He thought that studying technology would help him pivot into any field, including medicine, and that it would also earn him money that he could use to fund his own research lab. He feared that in academics, “I would be subject to the mercy of the grant givers,” he says. “And much of what I do would need to follow the general paradigm. And I wanted to focus on things that get into consumers’ hands—and also more ambitious projects.”
After university, Zhavoronkov worked in the computer chip industry, first in Canada, then in Germany. He ended up making some money when one company he worked at, ATI, was acquired by AMD. After three years in the industry, Zhavoronkov decided it was time to go back to school. He simultaneously did a physics PhD at Moscow State University and a master’s degree in biotechnology at Johns Hopkins University. He chose Johns Hopkins because of a lab there that studied long-lived wasps, science he thought might translate to humans.
Zhavoronkov’s first project at Johns Hopkins was to develop the nonprofit International Aging Research Portfolio, still active at AgingPortfolio.org. He gathered documents from several databases and tracked research projects related to aging from grants to publications, patents, and clinical trials. Researchers can now look at the efficacy of $2 trillion spent in the area. At Insilico, Zhavoronkov uses the aging portfolio to direct attention toward promising but underfunded areas. He founded the company after six years advising biotech firms including San Diego-based Sequenom, founded by Charles Cantor, the former principal scientist of the Human Genome Project and a current Insilico advisor.
AI all the way
Drug design is kind of a crapshoot. In a press release about their fibrosis drug candidate, Insilico declared, “We are proud to contribute to the transition of the pharmaceutical industry from an oftentimes unpredictable, serendipity-driven and organizationally disconnected drug discovery environment, which resembles a drug hunter’s craftsmanship, towards a more predictable AI-driven process—essentially, a move towards industrialized drug discovery.”
As Zhavoronkov explains it, there are roughly ten thousand diseases (though the actual number is disputed and varies depending on the definition). On top of that, there are sixty thousand possible biological targets—molecules, such as proteins, RNA, or DNA, whose functioning one might want to enhance or inhibit. Multiply ten thousand by sixty thousand and you have, well, a lot of hay hiding a few needles. To find those needles, pharmaceutical companies often sift through existing research, looking for promising glints. Perhaps someone saw an association between a disease and a protein, so the company explores that association. But if the association was already reported, it means the target isn’t novel, so it’s hard to patent a treatment. It also means someone may have already tried and failed to work with that target.
Insilico’s solution is a software pipeline they call Pharma.AI. They started by creating a tool called PandaOmics that discovers targets. Zhavoronkov shows me a demo over Zoom. Let’s say you’re going after colorectal cancer. PandaOmics digests the documents collected for the aging portfolio. For this cancer, that includes 16,000 grants, 5,000 clinical trials, and 100,000 publications. “For a human to read that, an entire lifetime would not be enough,” he says.
PandaOmics looks at the datasets collected through those studies and compares people diagnosed with colorectal cancer against people without it. It then lists genes whose expression, or production of proteins, differs between them. The main screen displays a grid of colored squares. Each of 20 rows represents one gene. Each of more than 20 columns represents a score of that gene’s relevance to the disease based on a different AI algorithm. Darker shades of green stand for a stronger link.
The first set of columns represents “omics” scores. Omics refers to the study of the entirety of a set of biomolecular mechanisms. For instance, genomics looks at the whole set of genes in an organism or a species. Transcriptomics looks at the degree to which those genes are transcribed. After the omics scores come text-based scores, based on the processing of basic linguistic patterns, such as the co-occurrence of terms, in the reports. Then there are financial and KOL (key opinion leader) scores, based on stock reports and the number of credible scientists working on the disease-target interaction. Columns further to the right indicate whether the target is accessible by small molecules, based on published structures and other data; whether there is any existing toxicity data in humans or animals showing it’s safe to interfere with; and how novel such treatments would be.
Insilico uses the same platform to uncover dual-purpose therapeutics—those that would not just treat disease but increase longevity.
Combining the columns, Zhavoronkov says, “you want to find this balance between novelty and confidence.”
“But if you want to go one step further, and do it the Insilico way, we use the same platform to uncover dual-purpose therapeutics,” Zhavoronkov says—those that would not just treat disease but increase longevity. So they might add more columns indicating effects on lifespan. In a paper published this year, they used PandaOmics to identify 145 targets associated with what the field calls the nine hallmarks of aging, including such phenomena as mitochondrial dysfunction and altered intercellular communications. “It’s actually a pretty breakthrough paper, if you think about it,” Zhavoronkov says, “because we did the work that many other people should be doing.” (He frequently acts as the company’s hype man.)
Insilico hasn’t published detailed papers on how PandaOmics’ scoring algorithms learn or operate; Zhavoronkov says it would be “stupid” to reveal their recipe. In that sense, the models are black boxes. (Zhavoronkov objects to the characterization of PandaOmics overall as a black box, because users can see individual algorithms’ scores and zoom in on individual papers.)
In 2016, Insilico tackled the next step in the pipeline, designing drugs to inhibit those targets. What they produced is a tool called Chemistry42. Input the chemical structure of a target protein, and it will propose a catalog of molecules that fit it snugly and deactivate it. The software resembles two other kinds of AI software. First, it’s generative, similar to deepfake algorithms, which can produce images of faces that look real. Here, it generates chemical structures that look plausibly like real drugs.
Second, it uses reinforcement learning, the kind of AI that helps software learn to play games or control robots through trial and error. Chemistry42 explores the abstract space of potential molecular structures and is rewarded not for beating an opponent but for creating molecules that other algorithms score highly for predicted safety, potency, stability, or other factors. Zhavoronkov says that nine of the top 30 pharmaceutical companies have licensed Chemistry42. They can add their own datasets or scoring algorithms to tweak the design process.
The third piece of Pharma.AI, InClinico, predicts the outcomes of clinical trials, using some of the same scoring algorithms as PandaOmics. Obviously, it will not have data related to any novel molecules, but it might have data related to the target. Its biggest customers are banks and hedge funds betting on the success or failure of pharmaceutical companies. On November 14, Insilico will reveal its latest version.
Licensing Pharma.AI to other companies is not a huge source of revenue, so Insilico’s main business is developing drugs internally (or partnering with other drug designers—this week they announced a collaboration with Sanofi worth up to $1.2 billion). In addition to the fibrosis treatment, they’re exploring about 30 molecules for cancer, inflammation, brain disorders, and COVID-19. They publish much of their data, but they have kept the fibrosis target a secret because it’s completely novel, Zhavoronkov says. When asked why they don’t keep all of their work secret, he says they publish in part to popularize and validate their pipeline.
“Insilico is extremely credible, scientifically,” says Evelyne Bischof, a physician at Shanghai University of Medicine and Health Sciences and a research collaborator of Zhavoronkov’s. “Insilico publishes a paper every month or something. That’s unusual for a company to do.”
Aside from Pharma.AI, Insilico also uses another tool in-house: generative-adversarial networks, or GANs. They use these AI algorithms to take a real or synthetic medical profile and artificially age the person, the way others have used GANs to show what the faces of celebrities would look like with a different age or gender. It’s “basically AI imagination,” Zhavoronkov says. They can perturb genes in an individual, fast-forward, and see how it plays out in the person’s health. It’s another way to discover targets for slowing aging.
Michael Levitt, a structural biologist at Stanford who won a Nobel Prize in chemistry in 2013, sits on Insilico’s scientific advisory board. He says Zhavoronkov’s most defining characteristic is the span of his knowledge. “He’s very broad, which is why he can have a company that goes all the way from finding a target to having a successful drug and understand the whole process of regulation of clinical trials. And he also knows which way one can accelerate each of these things.” He adds, “the pipeline, it’s all in his head.”
Zhavoronkov also motivates those around him. “When you meet Alex, it’s immediately obvious to you the gravitas of his vision,” says Alex Aliper, Insilico’s president and co-founder. “His energy is so enormous that it molds different types of people together. You have to bring biologists, chemists, clinical development specialists, drug hunters, AI engineers, and software developers together to create Insilico.”
The strength of Zhavoronkov also makes Insilico somewhat vulnerable, however. “I worry sometimes that Alex is so central to this,” Levitt says. The leader can’t be everywhere at once, and might someday not be there at all. (Zhavoronkov does have a co-CEO, Ren Feng, who leads drug discovery while Zhavoronkov leads software development.)
What does he do to stay in top form? “I don’t try to follow any specific longevity regimen,” he says. “I’m not one of those crazy people who just try to optimize” everything, he told me. “I try not to overeat.” He does 20 chin-ups in the morning. (He’s often in tight-fitting Insilico-branded polo shirts that show off his muscles.) Zhavoronkov says lifestyle changes can increase longevity but not nearly as much as drugs could someday.
But he does more than most. He claims he can do 100 consecutive pushups. He tries to follow a ketogenic diet, which increases protein and reduces glucose. Every few weeks he fasts for a day or two. He also takes a number of drugs and supplements that some people believe increase human lifespan but that lack conclusive clinical evidence and that also could pose health risks.
“The history of computational chemistry and computational drug discovery is a cratered plain littered with smoldering piles of overpromises.”
We can’t know if they’re extending Zhavoronkov’s life, but a blood test once indicated he had the transcriptome of someone four years younger. And on a recent Zoom call, an online tool at Haut.AI estimated his age based on his face. Despite his all-nighter, it pegged him at 38. (He’s 43.)
Besides Zhavoronkov’s central command, the company also faces other challenges. There’s very little data on certain rare diseases, Aliper says. And, Kristen Fortney says, “It continues to be very hard to predict how a drug will behave in patients. Insilico is offsetting that by having lots of programs and multiple ways to win.”
“The history of computational chemistry and computational drug discovery is a cratered plain littered with smoldering piles of overpromises from folks who got carried away,” says Derek Lowe, a drug discovery chemist at Novartis. The biggest pain points are not computational limitations, he says, but surprises in the clinic. He hasn’t seen enough results from Insilico or their competitors to know what AI can overcome. But, he says, “I’m a long-term optimist,” and “Insilico is certainly one of the leading companies in this space.”
Science above politics
In Zhavoronkov’s longevity pledge, he lists several initiatives he’ll support with his time and money. One is education. He’s developed free online courses for physicians, available at the Longevity Education Hub. If doctors understand the aging process, they might make different decisions about how to treat their patients, for instance trying new medicines or treating people based on their biological rather than chronological age. Additional courses are coming for investors in longevity and for veterinarians.
He also organizes the ARDD conference. He coaches researchers. He works on outreach, for instance writing regularly for Forbes. He advises governments and organizations. And he supports fledgling scientists. His Inspire Longevity program invites high school students to ARDD and gives them access to mentors and tools. This year, a group of students, working on their own with occasional guidance from Insilico, discovered novel targets for treating glioblastoma, one of the most aggressive and deadly forms of brain cancer. PandaOmics is “so simple, kids can use it,” Zhavoronkov says. “So if you direct them into longevity from the get-go, and they are naturally smart, and they get the AI tools, they can probably outperform pretty much anybody out there. And they can outperform me,” he adds. “I have a very weird career track.”
The pledge also lists several core interests. There’s the development of a pipeline for drug discovery (Pharma.AI). There’s the discovery of aging clocks, methods for assessing biological age (in 2020 Insilico spun off Deep Longevity to work on such clocks). Another is cryobiology. He spends his Sundays working out how to freeze and unfreeze biological material. “I think that this technology, just like deep learning, can be used for everything. It would enable entire new industries. I think that it could be bigger than the internet.”
It could be used for food or organs or emergency medicine or hibernation until certain diseases are cured. Or what he calls cryo-tourism. You might sleep 50 weeks a year and then check out the current year for two weeks. Like longevity, it lets you see more stuff. “One way is to create longer life,” Zhavoronkov says. “And another way is to just stretch this life.” He’s not doing experiments just yet; he’s still reading about the physics of gases at different pressures and temperatures. “He’s always a source of inspiration,” Bischof says, “showing us that he’s learning, and quite a lot. Okay, different things, sometimes very abstract for us, but learning.”
“I don’t see why we shouldn’t democratize drug discovery. And that’s the long-term vision.”
Another interest is automating physical aspects of the drug discovery pipeline. In December or January, Insilico plans to unveil an AI-run robotics lab, in which machines will culture cells, run experiments on them, image them, and feed all the data back into Pharma.AI, which will then inform further experiments. Bischof calls it his greatest project. “That’s, as you can imagine, a huge undertaking, and ‘no humans allowed,’ as he says.” Zhavoronkov foresees a day when hospitals have such facilities for personalized medicine based on patients’ biological samples. The system would identify promising FDA-approved or more experimental drugs, or discover new ones that have good predicted safety profiles and could be synthesized elsewhere.
Hospital-based labs could also feed (anonymized) data into global research efforts. Zhavoronkov notes that Basel, a smallish city hosting the headquarters of Roche and Novartis, is responsible for a disproportionate amount of the world’s pharmaceutical R&D. “We want to have Basel in a lab,” he says. “You put Basel into Riyadh or Abu Dhabi, and you get promising targets. And we will enable emerging countries, from scratch, to be able to kickstart their drug discovery programs. Why not Nigeria, right? Why not Rwanda?” He goes on, “I don’t see why we shouldn’t democratize drug discovery. And that’s the long-term vision.”
The robotics lab is in Suzhou, China, where Zhavoronkov spends much of his time (though he’s on the road giving talks so much that he doesn’t have a home and lives in hotels). Promoting international collaboration is part of his pledge, and he feels China is misunderstood. “The media is portraying China in a negative way, pretty much ubiquitously,” he says. He notes that they’ve lifted 800 million people out of poverty and with the COVID-19 drop in U.S. life expectancy, people in China now have a longer average lifespan than the United States, by some estimates. And he claims they’re business friendly. “I think that the recent trends in limiting cooperation in technology, it’s a disaster,” he says. “Every politician that does that should be out immediately. A drug discovered in China will be saving lives.”
Bischof says Zhavoronkov puts science above politics, which makes him welcome in places like Saudi Arabia and China, a country that’s moving quickly in medicine and technology. She notes that Insilico has a great reputation there. “That’s not easy,” she says. “And Alex will not say this, but Chinese leadership on the highest levels values both the company and Alex very, very much. That was surely a part of the rapid development of Insilico in the past two years.”
A new religion
Zhavoronkov considers himself part of the effective altruism movement, which applies bean-counting to ethics and philanthropy. A popular EA idea is called earning to give, in which people take high-paying though often unrewarding jobs so they can donate the proceeds. But Zhavoronkov says that’s not enough. He points to the two trillion dollars governments have spent on biomedical research in the past 35 years that has produced meager results for longevity. He hopes to set an example with his “dual-purpose career,” making money in an industry (information technology) that also gives him applicable skills. “This way, I could be pretty close to science,” he says. “And of course, you make the money to go after your big dream. That’s why I am preaching right now the same concept to many people in IT.”
“You should not worry about robots becoming conscious and killing you in fifty years when aging is going to kill you for sure.”
And of the causes to support, he sees healthy longevity as the most important (after preventing nuclear war, but he has no feasible method to depose Putin). “You should not give too much priority to low-probability existential threats like evil AI or self-replicating robots,” as some effective altruists focus on, he says. For instance, “you should not worry about robots becoming conscious and killing you in fifty years when aging is going to kill you for sure.”
Zhavoronkov’s prime metric of effectiveness is quality-adjusted life years, or QALYs. He has no wife or children and says having a family would just be “a big distraction at this point in time.” Do the math. “If you reproduce, you make, I would say 100, 300, 400 quality-adjusted life years,” he says. “If you make a drug that benefits everybody on the planet, and adds one year of life for everyone, you make eight billion quality-adjusted life years.”
(In his pledge, he adds, “I calculated that I can probably spend 6–7 hours a week on my personal life and I don’t want to disappoint anyone by being a bad husband or a bad father. I will not marry until I meet someone as dedicated to the cause as I am.”)
He also thinks the concept of work-life balance is “oversold.” His hobbies include writing papers, writing articles, and teaching—things that some people might consider work. “An interesting thing about Alex that definitely influenced me a lot,” Bischof says, “is that he prioritizes work over physiology, so sleep and stuff like that. And only that will bring success.” Aliper says Zhavoronkov always has his eyes on the overall mission. “It’s sort of like how Elon Musk judges every decision on whether it brings him closer to Mars.”
I asked Zhavoronkov if he saw a parallel between his previous religious devotion and his current commitment to increasing longevity. “Well, we could call it religious dedication, yeah,” he said. “But it’s not a religion, it’s just this calculated, logical thing to do. Maximizing quality-adjusted life years, that’s the only commandment of my longevity religion.”