“The pharmaceutical industry is pretty unusual in that 95 per cent of what it does fails,” Jackie Hunter, director at bioinformatics startup BenevolentAI, says. “For an evidence-based industry, we don’t actually use a lot of it. There’s a new paper out every 30 seconds. There is a huge amount of information out there that just isn’t being used for discovery and development of new drugs.” But maybe that’s about to change.
To tackle this problem, BenevolentAI uses artificial intelligence to mine and analyse biomedical information, from clinical trails data to academic papers. Following this approach, BenevolentAI can, for instance, identify molecules that have failed in clinical trials and predict how these same compounds can instead be more efficient targeting other diseases.
“Many compounds go into clinic testing in volunteers for a particular disease and don’t work because the hypothesis was wrong,” Hunter says. “Our system provides unbiased hypotheses. We can take that molecule and find a new disease to target and go straight to a clinical trial without repeating all the previous early testing that’s been done.”
BenevolentAI’s first clinical trial will begin this year in the USA and Europe, targeting excessive daytime sleepiness in Parkinson’s disease. “It’s a new indication for this particular molecule.” Hunter says. “We are really showing that that work has not been in vain that, actually, we can find a new use to this molecule in a different disease.”
Benevolent AI can also use the predictive power of its AI algorithm to design new molecules, extracting new hypothesis based on a knowledge graph composed of over a billion relationships between genes, targets, diseases, proteins and drugs. “When the periodic table of the elements was generated, there were gaps in that table where you know elements had to exist, but they hadn’t been discovered,” Hunter says. “We use our knowledge graph like that: what relationship should be present but are not yet known?”
The company is currently researching new potential therapies for ALS and has one clinical trial planned for next year with a new molecule designed in-house. “When we ran the query about ALS, our system came back with a rank list of hypotheses,” says Hunter. “The system isn’t 100% right all the time, so the end result is really a combination of the platform and the insight of the scientists.” In the case of ALS, BenevolentAI picked five new compounds and tested on cell cultures with motor neurone disease. “One failed the test, three worked as well as current gold standard and one worked exceptionally well,” Hunter says. “We have an end-to-end organisation capable to doing very basic, exploratory science right the way through clinical trials.”
The London-based startup, founded in 2013 by biotech entrepreneur Ken Mulvany, currently employs 90 people in the UK and US and has raised £109.5m. In September 2016, BenevolentAI hired Jerome Pesenti, former VP for IBM’s Watson.
“The way the industry is set up is very traditional. Drug discovery hasn’t really changed much,” Hunter says. “If we can show that our approach works in a really difficult, complex sector like the pharmaceutical industry and human biomedicine, then the potential in other less complex industries, like agrotech and veterinary medicine, is much greater. Any research intensive industry could benefit from this technology.”