In June 1993, David Brown was exasperated. For eight years he had been developing a drug that was supposed to treat angina, but early clinical trials showed it wasn’t having enough of an effect to make it into a commercial success. His employer, Pfizer, gave him three months to turn the project around. A few weeks later, Brown heard some unusual news from a group of Welsh men involved in a clinical trial of the drug: they were experiencing more erections than usual. Realising that it could have a huge blockbuster on its hands, Pfizer switched its clinical trials to focus on erectile dysfunction. “It went from dead to number one in the Pfizer portfolio in the space of two weeks,” Brown says. Approved in the United States in 1998, Viagra went on to sell more than $400 million (£237m) in its first three months alone.
Now Brown wants to repeat this trick to find new treatments for rare diseases with his startup, Healx. Co-founded in 2014 with Tim Guilliams, the Cambridge-based startup uses artificial intelligence to identify existing drugs that have already been through clinical trials, and then repurposes them as treatments for rare diseases. Over 400 million people worldwide live with such diseases, but because each one affects a small number of people, pharmaceutical firms don’t have the financial incentive to develop new drugs to treat them. “The quickest and safest and most cost-effective way of inventing a new drug is to start from an old drug,” says Guilliams, Healx’s CEO.
The idea for Healx came from a conversation Guilliams and Brown had in a Cambridge pub with Nick Sireau, whose two sons have a rare genetic disease called alkaptonuria. Although there are around 7,000 rare diseases, 95 per cent of them lack any treatments. When treatments are developed, they can be extremely expensive. Zolgensma, a drug approved in the US to treat spinal muscular atrophy, costs $2.1 million (£1.5m) for a single one-time infusion. The huge failure rate when it comes to developing new drugs drives up the cost. According to one estimate, only four per cent of drugs developed manage to make it all the way to approval.
Guillams and Brown want to massively improve these odds. At the heart of Healx’s approach is Healnet: an AI platform that finds links between drugs and diseases. Using natural language processing, the platform mines data from clinical trials, patents, electronic health records, genomic datasets and scientific papers to find drugs, or combinations of drugs, that match a particular disease profile. At the same time, Healx scientists are creating and curating new datasets to fill in knowledge gaps about new diseases. “You put that human mind alongside a computer, and the computer helps the human mind be more effective, and the human mind helps the computer be more effective,” says Brown.
One rare disease Healx is working on is called fragile X syndrome; one of the most common causes of inherited learning disability. In 2017, it started identifying a number of repurposed drug combinations that showed promise in mouse trials, and now has plans to start clinical trials in 2021. The firm is also searching for treatments for other rare diseases, including Pitt-Hopkins syndrome, Facioscapulohumeral muscular dystrophy and some rare cancers. Eventually, Brown hopes that Healx’s work won’t just find new treatments for rare diseases, but completely overhaul how the industry finds new drugs. “I think we can transform the whole drug discovery process as well,” he says.
