Artificial intelligence as the co-pilot for drug discovery
There are more molecules that could conceivably be candidate drugs than there are stars in the universe. How can we ever efficiently identify those molecules? Professor of AI and Medicinal Chemistry, Gerard van Westen: ‘I’m going to use artificial intelligence as the co-pilot to make an automated search.’ He will give his inaugural lecture on 26 May.
It takes around fifteen years and more than a billion euros to develop an effective drug. ‘You discover candidate drugs or produce them in the lab. First you test them in vitro, then in an animal model and finally in humans.’ In most cases, the molecule is eventually rejected.
Death due to side effects
However, if all the tests have given you a potentially effective drug, it still needs to be stable, safe, soluble and suitable for large-scale production. Among other things. Van Westen: ‘All the factors you have to take into account are almost beyond the comprehension of the human mind. A brilliant example is the withdrawal of the anti-inflammatory Vioxx, an ibuprofen-like drug that turned out to have the side effect of heart rhythm disorders. It blocked the hERG channel in the heart. Since that became known, every candidate drug has to be checked for its effect on that channel.’
The human mind can enlist the help of artificial intelligence at every step on the way to a new drug. It begins even when thinking about which of the molecules in the gigantic chemical universe you want to produce and test in the lab. ‘Our search space is limited by time and money. But we can refine the search through that universe a little with the tools that I’m making: for example, an idea generator that proposes 10,000 ideas, linked to spam filters that exclude the bad ones. You’re then left with around ten molecules that are easiest to make in the lab and have the best chance of success as a drug.’
A drug like an F-16
In his presentations, Van Westen often shows a picture of the F-16 fighter jet to illustrate his ambition. ‘That plane was designed entirely by computer. When they first put it on the runway, the designers knew it would go up in the air. But they didn’t know how fast it could fly and how sharply it could turn. With drugs it’s actually the other way round. Out of the drugs that arrive on our lab research runway, many of them simply don’t work. We know more about aerodynamics than about our organism.’ Consequently, it’s not yet possible to create candidate drugs as efficiently as the F-16.
‘Scientists who use artificial intelligence will replace the ones that don’t.’
Van Westen is a member of a consortium that aims to bring us closer to the F-16 ambition, with a grant from the Dutch Research Agenda (NWA), in the Virtual Human 4 Safety project. ‘It would ultimately be fantastic to create a digital patient for drug safety testing without lab animals.’ This would be possible with algorithms that have learnt from all the knowledge that exists about molecules and their reactions with all the proteins in the human body. ‘Some day it will happen.’
Virtual human
Artificial intelligence has actually been receiving a lot of bad press recently. Might we not be taken over by this virtual human? Van Westen isn’t afraid of this. ‘Artificial intelligence won’t replace scientists, but scientists who use artificial intelligence will replace the ones that don’t.’
Text: Rianne Lindhout
Photo: Patricia Nauta