ChatGPT and other large language models have seen a surge in popularity in recent months, transforming firms like OpenAI into multi-billion dollar businesses and amassing hundreds of millions of users worldwide, employing the tool to summarise reports, answer queries or write people’s homework.
But in a small office block wedged in between St Pancras Station and Google’s giant new London headquarters, a team of researchers is exploring another purpose for large language models: to understand the language of DNA.
Their work in building up a picture of the structure of human DNA and how it influences medical outcomes, combined with a range of other AI-powered tools being built, is set to rapidly transform the world of pharma, including improved diagnoses, better treatment and faster drug development.
“AI and machine learning has been around for a long time but what’s really interesting now is this explosion of data in biology and medicine,” said Kim Branson, the global head of artificial intelligence at GSK who manages the team of about 50 at King’s Cross.
“We can take a tissue and see each individual cell and measure all these things in amazingly fine detail. You can take a sequence of DNA and figure out which gene to turn on or off to increase or decrease a mutation.
“You can basically glue things together that you couldn’t glue together before.”
The work of the AI team at GSK has already shown promise since the unit was set up in 2019. Its efforts were pivotal in the development of the firm’s new drug bepirovirsen, a potentially transformative treatment for hepatitis B, which claims the lives of nearly a million people each year.
GSK’s use of AI is steadily reaching every arm of its business as its AI unit grows at pace, with outposts in San Francisco and Boston. The firm recently recruited two new AI specialists to its board, in signs the technology will become central to its work in future.
“We’re at work in both the early discovery aspect of things but also in the later-stage clinical aspects,” said Branson. “We’re able to use data to build a model of who is most likely to respond to treatment and work it out ahead of time.”
According to insights firm Deep Pharma Intelligence, investment into AI-driven drug discovery firms tripled over four years to just under £20 billion in 2022. And the UK is punching above its weight on investment in AI for big pharma.
Of the roughly 800 AI companies involved in drug discovery, around one in 10 are based here — almost two-thirds the number in the EU as a whole — while around one in 12 of the 1900 global pharma AI investors are also UK-based. Pharma giants are racing to snap up AI start-ups, with AstraZeneca topping the leaderboard after signing 27 separate deals.
On average, it takes 10 years to bring a new drug to market. Undoubtedly, AI is helping cut that time for firms like GSK. But without being able to bypass clinical trials and regulatory approval, it can only go so far.
The real difference artificial intelligence can make to the pharmaceutical industry is in the economics. According to the Tufts Center for the Study of Drug Development, the cost of developing a new prescription drug that gains market approval is approximately $2.6 billion (£2.1 billion). That figure could be slashed massively if drug discovery and pre-clinical development becomes automated. That could mean drug costs falling, and pricy treatments the NHS considers uneconomic today could become affordable in the years ahead.
“It’s a speed and quality thing but really… the bigger picture is actually the quality effect,” Branson said.
“I don’t want to be wrong faster because that’s just going to waste more money. [But] by integrating genetic data with other types of data, you’re less likely to fail in clinical trial.”
