Where Next for AI in Drug Discovery?

Adrian Tombling explains why drug developers need AI

WITH the cost of bringing a new drug to market now an average US$2.6bn1 and one-in-ten drug candidates failing to make it to market despite successfully completing Phase I trials2, it is no wonder that pharmaceutical companies have seized on the unparalleled data-processing potential of artificial intelligence (AI) systems.

Their use in identifying compounds, some of which may have completed clinical trials already, that could be re-purposed to treat alternative diseases quickly and comparatively cheaply, is well documented. But as research scientists are beginning to find, AI systems are capable of achieving so much more.

The re-purposing opportunity

The potential applications of AI in drug discovery are almost endless, but one of the main areas of focus to date has been repurposing existing drugs. Typically, this involves finding new uses for drugs that have already attained market and regulatory approvals for the treatment of a specific disease. Using AI technology to analyse existing research data, which might include information from clinical trials and other patient data, it is possible to determine whether a drug molecule will bind to other specific targets. This information can then be used to help understand how effective the drug might be at certain dosages or when treating new patient groups. When used in this way, AI systems can help to identify re-purposing opportunities more quickly and efficiently than would be possible using traditional scientific research methods.

In fact, AI systems have the potential to provide a more definitive view of a re-purposed drug’s potential than would otherwise be possible. AI systems can determine whether a drug compound binds to multiple targets, and whether by binding to such targets the drug has the potential to treat diseases associated with one or more of the targets. Unlike research activities undertaken by humans, the analysis provided by AI systems is guaranteed to be objective as it is based on patterns derived from known data sources.

Pooling data improves outcomes

Of course, the flaw in this particular use of AI systems is that the analysis provided is only as good as the quality of the datasets in use. For this reason, the pharmaceutical industry is increasingly seeking to collaborate, to pool data and use it to train algorithms through a process of machine learning. A recent initiative involving no fewer than ten pharmaceutical companies – including GSK, Johnson & Johnson and AstraZeneca – known as the Melloddy Project, is using a novel blockchain system to store data on a secure ledger, whilst protecting the trade secrets of individual companies.

Growing use of AI in identifying re-purposing opportunities could help to find cures for other diseases that affect smaller sub-groups of the population or people in third-world countries, where funding for drug discovery programmes is in short supply

There are many examples of early success in the area of re-purposed drugs. An innovative AI technology developed by AI specialist Atomwise is using deep neural networks to aid drug discovery by analysing simulations of molecules in order to reduce the time that research scientists need to spend synthesising and testing compounds. In a collaboration with IBM and the University of Toronto in 2015, the company utilised its AtomNet technology to analyse and predict the molecules that could potentially bind to a particular glycoprotein in order to find a treatment for Ebola virus infections that had caused the death of over 11,000 people in Africa and some other parts of the world. More recently, Merck has been using AtomNet to scan its existing medicines in order to identify any opportunities to re-purpose them to fight existing or upcoming diseases.

It is hoped that AI technology could deliver benefits in areas of drug discovery that typically attract less funding. Big pharmaceutical companies are focussed on finding a cure for the most prevalent and debilitating diseases, such as cancer and Alzheimer’s, and this is unlikely to change. Growing use of AI in identifying re-purposing opportunities could help to find cures for other diseases that affect smaller sub-groups of the population or people in third-world countries, where funding for drug discovery programmes is in short supply. Pharnext is an example of a company deploying machine learning to identify compounds for the treatment of rare disorders, and it currently has a compound going through clinical trials for the treatment of a rare neurodegenerative condition called

Charcot-Marie-Tooth disease. As the original compound has already been shown to be safe, the re-purposed version can make it to market more quickly, as certain aspects of the clinical trials can be avoided.