London-based biotech business BenevolentBio is on a mission to overcome the failure of the drug industry to innovate on drug discovery. Andrew Huddart explains and talks to its chief executive, Jackie Hunter.
The tools of big data are helping to reshape the industrial landscape of the 21st century. They are helping make entire economic activities obsolete and helping give rise to new ones. The area of health and drug discovery in particular is far from excluded from this change.
One London company, BenevolentBio, is on a mission to use data science to shake up one of largest and most capital-consuming analogue industries – drug discovery.
BenevolentBio, founded in 2013, puts artificial intelligence and high-performance computing in the hands of its team of experts in life sciences, maths and data to disrupt the entire pharmaceutical research and development value chain – from target selection through compound development to safety and efficacy trials.
Such expert-augmenting technology will, it says, enable “the previously impossible”, and perhaps rescue a pharma industry plagued with R&D inefficiencies and struggling against headwinds of cost and regulation.
The world’s top 15 pharma companies spend between $3bn and $9bn a year each on research. Altogether nearly $87bn (£67bn) is invested primarily by US and European firms (Takeda of Japan being the exception on this list).
But despite this approvals of new candidate compounds are falling, and when manufacturers raise prices of existing drugs to fill income gaps they inevitably face public outcry from the media, politicians, and regulators.
BenevolentBio chief executive Jackie Hunter – a former pharma insider – believes that AI could reverse what she calls “the failure of innovation” that seems to plague the global drug industry.
Hunter recognises that price hikes understandably get attention, but sees them as a side effect. “It’s the fact that there are high prices and there isn’t the level of innovation that people expect, given the level of investment,” she says. “This is why I don’t think the industry is sustainable. Because that level of failure just isn’t sustainable.”
Hunter believes Benevolent can reverse the trend by looking differently at the raw materials of drug discovery. Instead of adding new laboratory science, Benevolent has designed an engine it calls ‘knowledge graph’, to extract deep learning from the intelligent analysis of known scientific information.
The focus, Hunter explains, is on revisiting the potential of drugs that got stuck in development (so-called ‘halted’ or ‘stalled’ assets).
The method sounds simple. Machines do the repetitive work of reading the latest scientific research, daily, applying AI learning to make associations between biological entities based on credible evidence, and then present the ‘graph’ to researchers to analyse and generate their hypotheses.
Simple, that is, until you realise the numbers involved. There are 2,300 known drugs and about 1021 drug-like molecules; around 400 identified drug targets, and 19,000 genes in the human genome; plus over 13,800 known diseases, of which about 5,000 are classified as treated and 7,000 ‘rare’.
This adds up to hundreds of millions of associations, and demands a very big computer. For Benevolent, the choice was the Nvidia DGX-1 – dubbed by its maker “the deep learning supercomputer in a box” – carrying a £100,000 price tag, plus ongoing support.
“For what it does, and the time it saves, it’s amazing,” says Hunter. “Ten years ago we wouldn’t have been able to do this because we wouldn’t have had the computing power, nor the storage capacity.”
As ever with science, the company’s big test will be proof – how well its digital hypotheses, and in-silico testing or validation of novel biochemistry – translate into cell-based and then clinical validity. Hunter promotes the benefit of slashing early R&D time and, alongside that, costs.
So Benevolent has a business strategy of alliances and partnerships with specialists to close the clinical gaps.
Rejecting the idea of selling its biomedical AI “as a service” to pharma, the company keeps its tech private and teams up on a project or venture basis only with research institutions, charities or niche pharma firms around specific diseases or targets.
Initially, Benevolent will manage without its own ‘wet’ laboratories, sticking to its current open-plan office. It envisages that, within a few years, it will control entire drug development lines itself for some of its major programmes, and contract specialist manufacturers for final production.
Why, then, the Benevolent name? After all, how can an artificial intelligence company be and remain benevolent?
Given the increasing strains around data science as big business, where the public and media have real concerns around responsibility and risk, the response is revealing.
Ken Mulvany, founder-director of the BenevolentAI holding company that includes BenevolentBio, offers it in a 700-word blog explaining the firm’s desire to work for science and public good.
Originally it had a more mundane name, Stratified Medical, which signaled its biological intent. However, Mulvany sees the now four-year-old group as developing and applying AI technology to enhance and accelerate wider scientific discovery.
“In essence we are hard wiring morality into the AI learning process so that the technology’s aim is implicitly to benefit society by improving quality of life (health) and protecting way of life (a better place to live),” he says.
For now, backed by a total equity raise of £68 million, led by UK patient capital investor Neil Woodford and foundations linked to mid-sized pharmaceutical firms, Benevolent is growing fast and making the most of its home in London’s Knowledge Quarter (the area in a one-mile radius around King’s Cross).
“It’s incredible, and that’s why if we expand we’re never going to move away from this area because it’s just where we want to be,” Hunter says.
Culture matters as much as scientific and technical resource, she says, and Benevolent is working to bring together biomedical talent and AI engineers and researchers.