Artificial intelligence: How AI could help discover new drugs billions of times faster – Lee Cronin

‘Artificial intelligence’ (AI) has become a mainstream term to describe anything that can perform tasks normally requiring human intelligence to complete, most typically areas such as visual perception, speech recognition and language translation.

The latest software and hardware advances in using large datasets to solve these types of well-defined problems have been incredible. But I question whether such input-led, largely prompted technologies, are genuinely ‘novel’ as we like to refer to anything ‘new’ in science.

I’d even go further and suggest AI is more of a mis-named branch of computer science and applied mathematics, because we really don’t yet have a rigorous definition of the ‘intelligence’ we find in living systems all around us. The trap we have fallen into follows an all-too-familiar model: acquire a vast amount of domain-specific data (for example, text, pictures, game positions or moves), then train a deep-learning system to create a model that can be used to classify, label, or generate findings.

Of course, there are lots of amazing tasks AI systems can carry out, albeit not yet for themselves. They can create realistic images and art from a description in language, for instance; or enable and then control self-driving cars; or allow optical and voice recognition, typically on security systems such as online banking access.

One problem which highlights their lack of independent ability is within drug discovery where the realm of molecular space for new drug molecules is bigger than outer space. This makes it very hard, time-consuming and eye-wateringly expensive to search for new drugs.

Right now, we rely on chance discoveries, but what if we could produce the equivalent of a chat-bot for new drugs? Could we use AI to search chemistry and biochemistry, record the data, and then use a type of drug-AI to suggest new potential molecules as drugs?

At the University of Glasgow, we are on the right track. In my own work, we have developed the ability to discover new drugs from within the vast realm of possibilities by using detailed data about molecule production fed into an AI model. This offers the possibility to virtually steer and control the process of molecule production in the hope to find a novel discovery, just like the AI control in a self-driving car navigating a map.

To achieve it, we developed a ‘Chemputer’ – a computer software or robot which can orchestrate virtual molecule production from scratch. This ability makes the process of searching for potential new drugs much faster by automating molecular building. So, rather than the prospect of new drug discovery being very low, the self-driving Chemputer programmes will enable us to explore chemical space billions of times faster.

Ai-Da Robot poses for pictures before making history as the first robot to speak at the House of Lords (Picture: Stefan Rousseau/PA Wire)

Once fully autonomous, it will be able to use AI to discover and then manufacture the drug too, giving humanity more rapid access to new drugs than ever before. At that point, the Chemputer will then become chemistry’s printing press.

Lee Cronin is the regius chair of chemistry in the School of Chemistry at the University of Glasgow, and a fellow of the Royal Society of Edinburgh. This article expresses his own views. The RSE is Scotland's National Academy, which brings great minds together to contribute to the social, cultural and economic well-being of Scotland. Find out more at rse.org.uk and @RoyalSocEd.