AI for drug discovery draws a $1 billion launch—and a lot of hope

The pursuit to use AI to discover new life-saving drugs got a big boost this week when a new company, Xaira Therapeutics, emerged from stealth with $1 billion in funding. The San Francisco-based firm aims to create AI models to develop new ways to connect biological targets and engineered molecules to human diseases. It’s led by Genentech’s former chief scientific officer and was incubated by Foresite Labs and Arch Venture Partners—the latter of which says Xaira is the largest initial funding commitment the firm has ever made in its 30 years of investing. 

Any funding round with a B—especially for an early-stage company—represents a big bet and a lot of hope. Aspirations are particularly high for the use of AI in drug discovery. While the upside of models that can create videos, music, and other creative works is still fuzzy (and in some ways even feel threatening), the societal benefits of breakthroughs in medicine that could help people live longer, healthier lives, and reduce suffering are crystal clear. It’s one of the use cases for AI I’ve been most interested in for this reason. Even my grandmother, who within the last year or so started hearing about AI on shows like “60 Minutes,” tells me that while she’s very worried about AI, she’s hopeful for what it could do to advance medical research and help people with conditions like Dementia. 

These pursuits are still in their very early days, but there are efforts underway, deals being made, and some signs of success. Data analysis firm StartUs Insights identified 463 AI startups working on drug discovery. Pharmaceutical behemoths are also pouring money into AI drug development efforts—just yesterday, Moderna announced a partnership with OpenAI to further incorporate AI into its drug discovery and development processes. Last year, BioNTech and Eli Lilly acquired AI drug discovery startups InstaDeep ($680 million) and XtalPi ($250 million), respectively. Big Tech is also diving in. Microsoft has partnered with Novo Nordisk and Google has released AI tools aimed specifically at helping pharmaceutical and biotech firms advance drug discovery and development. Last month also saw the first AI-targeted and AI-designed drug reach Phase II clinical trials—a drug developed by startup Insilico Medicine to treat idiopathic pulmonary fibrosis.

How AI can bolster drug discovery was also a hot topic at Fortune’s Brainstorm AI conferences in San Francisco late last year and last week in London. During a breakout session focused on AI’s role in drug discovery and clinical trials at the San Francisco conference, Amgen VP of digital, technology, and innovation Skott Skeller said this is a “hinge moment” for the biopharmaceutical industry that’s changing how drugs are developed.

“I often think, and someone once told me, that discovery and bringing drugs to market is all around identifying the cause of biology, poring over the chemistry, and then innovating like crazy to get that therapy to patients,” added Rory Kelleher, Nvidia’s global head of life sciences business development. “Think about it—AI can be implemented at every stage of that pipeline. And so we’re thinking about, how do we democratize those capabilities so that we can decrease the failure rate? If you decrease the failure rate, you reduce the cost and time it takes to market. And there are thousands of diseases with unmet needs.”

At this point, it’s important to note that AI isn’t a total silver bullet for drug discovery. While AI can drastically speed up finding promising compounds to make drugs, a lot of time-consuming and expensive steps of the process will still be required. For example, drugs will still need to be tested in wet labs—where they’re analyzed using physical samples—go through clinical trials, and gain FDA approvals. Still, AI has the potential to massively streamline that first step and make it possible to find targets that might not have been discovered otherwise. 

AI’s potential for health care doesn’t stop at drug discovery, however. Researchers are also examining how AI can bolster health care-focused robotics, analyze data to help make the best decisions for a patient’s care, aid in preventative health, and so many more facets of health care. As covered in Tuesday’s Eye on AI, startup Profluent AI this week demonstrated the first successful precision editing of the human genome with a programmable gene editor designed using AI. AI is also already being used widely in diagnosis, where multiple studies have shown it performs faster and more accurately than humans. One AI program for detecting breast cancer reliably interprets patient data 30 times faster than a human doctor and with 99% accuracy.

And with that, here’s more AI news.

Sage Lazzaro
sage.lazzaro@consultant.fortune.com
sagelazzaro.com

AI IN THE NEWS

Meta stock drops 16% after reporting spiking AI expenses. While the company saw revenue grow by 27% compared to the same period last year, Meta also reported it's increasing spending in order to accelerate its AI roadmap and that it will take years for the company to make money from the technology, Fortune’s Alexei Oreskovic reported. Meta’s capital expenditures for 2024 are now expected to range between $35 billion to $40 billion, an increase over a prior forecast of $30 billion to $37 billion that seems to have investors wary. “On the upside, once our new AI services reach scale, we have a strong track record of monetizing them effectively,” CEO Mark Zuckerberg said. Microsoft and Alphabet—whose earnings reports will also be all about AI—are set to report today after the market closes. 

The AI model Microsoft pulled last week because it didn’t get a safety test is still easily available. While the company deleted its WizardLM 2 model within hours of making it available after realizing it “accidentally missed” required safety testing prior to release, the move came too late. According to 404 Media, the model was downloaded during the brief window when it was available and quickly reuploaded to GitHub and Hugging Face, making it available for use anyway. “The Github and Hugging Face pages for WizardLM 2 are still down, but it is incredibly easy to find multiple instances of it being reuploaded on the same platforms,” reported 404 Media. The story is another example of some of the challenges that open-source AI models pose to safety and regulation.

Nvidia acquires AI infrastructure orchestration and management service Run:ai and AI model optimizer Deci. While terms of the deals haven’t been disclosed, TechCrunch reported Run.ai sold for $700 million. Founded in 2018, Tel Aviv-based Run:ai helps enterprises get the most out of their cloud computing resources when deploying AI. The move is an indication that Nvidia is looking to try to optimize highly energy-intensive data center computing loads amid rising concerns about the cost (with much of that being electricity) and carbon footprint of the generative AI boom. The Information also reported Nvidia is acquiring another Israeli AI infrastructure startup, Deci, that helps optimize AI models so they can run more efficiently on AI chips. You can read The Information story here. 

FORTUNE ON AI

Cargill leans on regenerative agriculture and generative AI to feed the planet —by John Kell

Move over, 'Black Mirror'; the BBC is showing AI can be a force for good in media —by Molly Flatt

The AI frenzy could fall flat as companies hoard chips without enough data centers to host them —by Aroosh Thillainathan (Commentary)

The race for human-AI interaction usage data is on—and the stakes are high —by Jeroen Van Hautte (Commentary)

AI CALENDAR

April 25: Microsoft and Alphabet report calendar Q1 earnings

May 7-11: International Conference on Learning Representations (ICLR) in Vienna

May 14: Google I/O conference

May 14: Stanford HAI’s RAISE Health Symposium 

May 21-23: Microsoft Build conference in Seattle

June 5: FedScoop’s FedTalks 2024 in Washington, D.C.

June 25-27: 2024 IEEE Conference on Artificial Intelligence in Singapore

July 15-17: Fortune Brainstorm Tech in Park City, Utah (Register here.)

July 30-31: Fortune Brainstorm AI in Singapore. (Register here.)

Aug. 12-14: Ai4 2024 in Las Vegas

BRAIN FOOD

Digital twins. Tech entrepreneur and investor Reed Hoffman yesterday shared a video to LinkedIn showing him interviewing—and later being interviewed by—an AI-generated version of himself, “his digital twin.” He said it was created with a custom GPT-based model trained on 20 years of his content including his podcast, interviews, and books he’s written, and it mostly looked and sounded like him. Reed introduced that he was going to challenge “ReedAI” with some questions to see what it can do, how good the content is, and how close it is to what he’d say. The dual interview, however, turned out to be a big softball. 

Reed asked the AI Reed to explain one of his books in one sentence for various audiences (the smartest person in the world, 5-year-olds, “StarkTrek” fans) and in different styles (Jerry from “Seinfeld”) He asked for advice on his LinkedIn page and what role government should play in regulating AI. Overall, he thought his digital twin used too many buzzwords. 

When AI Reed stepped into the interviewer role, human Reed launched into explaining why we need to embrace the AI “cognitive industrial revolution” and how it will elevate humanity. He also took issue with the fact that, according to him, the vast majority of people are only talking about the risks and not the benefits. Only once we realize how AI will elevate humanity can we bring in the risks, he said. 

The longer the video goes on, the less it feels like an experiment and the more it feels like a glaring ad for AI optimism in disguise. Reed is, of course, on the boards of Microsoft and Inflection AI and has invested heavily in many AI startups as a partner at Greylock Ventures. It’s a shame because I would’ve been really interested to see the model put to more tests.  Either way, seeing a person side-by-side with an AI-generated version of themselves was eerie. Cue the sci-fi comparisons.