Good morning. When it comes to revolutionizing drug discovery, AI hasn’t remotely lived up to the hype – more on that below, along with Jagmeet Singh’s plan for the carbon tax and The Globe’s gigantic fall cultural preview. But first:
Today’s headlines
- Air Canada calls for government intervention should its contract talks with pilots fail
- Ontario and Quebec will offer RSV shots for all newborns as other provinces stick to their current approach
- Russia says it will expel six British diplomats it accuses of spying and ‘subversive activities’
- TIFF cancels its screenings of the controversial Russians at War documentary, citing security concerns
A Deep Genomics lab technician at work.MARK SOMMERFELD/The New York Times News Service
Science
The billion-dollar AI drug gamble
Among the many promises of AI – that it will drive our cars, solve our cold cases, manage our money, write our newsletters – one of the most auspicious is its potential to keep us healthy. Making new drugs from scratch is a wildly time-consuming process. You have to choose a target for the drug (let’s go with a protein in the body), then create a molecule that meddles with that protein (say, shuts it down). Then you’ve got to ensure in a lab that the molecule shuts down the right protein (instead of some other vital one), then send it off to a clinical trial to show it is safe and effective in humans. That’s hardly guaranteed: 90 per cent of candidates fail in clinical trials. Little wonder it typically takes at least a decade and $1-billion to develop each new drug.
But by ditching fruitless molecules and predicting how better ones function in our bodies, AI is meant to speed drug discovery up and bring costs way down. Exciting! And also – as my colleagues Joe Castaldo and Sean Silcoff found – entirely unproven so far. In their new report, published this morning, they reveal there are zero AI-designed drugs in the market today.
Companies relying on AI to help with development, like Canadian biotech darling Deep Genomics, have suffered serious setbacks. “AI has really let us all down in the last decade when it comes to drug discovery,” Deep Genomics founder Brendan Frey told The Globe. “We’ve just seen failure after failure.”
Trial and error
So what happened here? Deep Genomics offers a handy case study. One major challenge is that experts in AI are rarely experts in pharmaceuticals as well, and you need those skilled drug developers to figure out the problems AI should pursue and validate its predictions in the lab. Deep Genomics didn’t initially have that experience on staff, and one veteran biotech entrepreneur, Clarissa Desjardins, told The Globe she heard Big Pharma is plucking people from food delivery service SkipTheDishes. (Her response: “Oh my God.”)
Deep Genomics founder Brendan Frey.Galit Rodan /The Globe and Mail
There’s also no AI model for the graveyard of most drug candidates: clinical trials. A new research paper followed 67 AI-designed molecules through this exacting process. The results found that while the molecules were quite successful in small Phase I safety trials, they failed at pretty much the same rate as other new drugs in larger Phase II effectiveness trials. Maybe scientists will someday build an AI replacement for the human body to test out drugs; we’re not there yet. That’s why Brian Bloom, CEO of Toronto health care–focused investment bank Bloom Burton & Co., isn’t particularly bullish on AI’s ability to save companies time. “In the 10-year journey of drug development, you may be saving five months,” he said.
Crunching the data
Data are essential to the smooth operation of an AI system, and the more complicated the system, the more data it needs. To predict protein structures, Google’s AlphaFold draws from the Protein Data Bank, a colossal resource of more than 200 million structures compiled over the past 50 years. That kind of training model is still rare, though, so companies like Deep Genomics have been left trying to build their own data or weed out the poor-quality stuff.
But wait, you ask – could this be a job for generative AI, which can be used for a whole range of tasks, even ones it wasn’t specifically trained to do? Great call, and exactly what Deep Genomics is now hoping. Last fall, the company announced a sort of ChatGPT for biology called BigRNA. A researcher feeds it a DNA or RNA sequence, and the AI system predicts, for example, which changes in DNA affect the production of a crucial protein and which ones switch on a gene at the wrong place or time. “BigRNA is connecting the dots between a DNA sequence and all of the molecular biology that actually leads to disease,” Frey told The Globe.
Will this lead to better, faster, cheaper AI-designed drugs? Right now, it’s hard to say. Ideally, the dot-connecting reveals why a drug is well-poised to work in humans, so researchers aren’t just throwing molecules at a wall. But whether Big Tech will finally make good on its promise to revolutionize health care – well, no AI system in the world can predict that yet.
The Shot
‘If I learn French, my chances to become a Canadian permanent resident are way higher’
Johann Loewenstein, a space engineering student from Colombia, spent three months in France learning the language.Yader Guzman/The Globe and Mail
After other paths to permanent residency proved difficult, hopeful immigrants to Canada are boning up on their French to increase their odds. Read more about the strategy here.
The Wrap
What else we’re following
At home: Jagmeet Singh signalled that the NDP plans to oppose the Liberals’ current carbon-tax plan, and that his party is working on its own climate policy ahead of the next election.
Abroad: Russian forces stepped up their attacks on the strategic city of Pokrovsk in eastern Ukraine, leaving residents without gas and their last source of drinking water.
Wrapping up: No huge surprise here, but Donald Trump said he won’t take part in another debate with Kamala Harris.
Settling up: Quebec just tabled a bill that would force businesses to calculate tips based on the price before tax, not after.
Settling in: In the market for some can’t-miss movies? Looking for a great new TV show? Maybe you’re a theatre buff or an art-gallery-goer or the kind of music lover who buys the album and the concert tickets in one fell swoop. Whatever your particular cultural itch, The Globe’s fall preview has you covered.