Leiden University logo.

nl en

Gerard van Westen: 'Our model predicts what candidate drugs do in your body'

He’s a fast and animated speaker, which is only logical because Gerard van Westen is driving an express train. His destination? A virtual human, consisting of algorithms that predict what an administered substance will do in the body. The train is already a long way down the line and the pharmaceutical industry is also on board.

A virtual human that can be used to predict whether a potential drug will have side-effects or be toxic as well as how it spreads through the body. That is the plan. With a bunch of partners from academia and business, but also the Netherlands National Institute for Public Health and the Environment (RIVM) and Proefdiervrij, an organisation against animal testing, Gerard van Westen has just heard the good news: under the leadership of Utrecht University, they have been awarded a Dutch Research Agenda grant to research such a model.

Gerard van Westen
Professor of Artificial Intelligence and Medicinal Chemistry Gerard van Westen works at the Leiden Academic Centre for Drug Discovery and the Leiden Institute of Chemistry.

Van Westen, since spring 2020 Professor of Artificial Intelligence and Medicinal Chemistry: ‘The part of the virtual human that I’m working on is the interaction between a candidate drug and proteins in our body.’ Drugs sometimes have to inhibit or, conversely, activate a certain protein, while leaving others alone. ‘Since the 1970s, 3.8 million high-quality unique measurements have been published about the interaction between a chemical substance – a candidate drug, for example – and a protein. These 3.8 million measurements form the dataset that we’ve used to train our model.’

'Which molecule would be able to do that? Our system can look for a match'

That model, fed with 3.9 million data points, can do two things. First, it is a library. Van Westen: ‘Imagine we want to block protein X because it mutates in a certain type of cancer and becomes too active. Which molecule would be able to do that? Our system can look for a match with a similar protein so that you know in which direction to look for a good drug.’ 

Swiss cheese with lots of holes

Second, the model can predict the interaction between a substance and a protein. ‘There are around a million relevant chemical structures in our dataset, and 5,500 proteins that you need to keep an eye on with candidate drugs. The product of this is 5.5 billion data points.’ Of these ‘only’ 3.8 million are known from measurements: see it as a Swiss Cheese with lots and lots of holes. 

In some parts there is quite a lot of cheese, but in others the hole is rather large, depending on the historical interest in certain proteins. Targeted measurements to fill the holes in strategic places would massively help the model make better predictions. ‘That is exactly what I want to do, but it’s difficult to get funding for it. We want to do it ourselves now, within the University’s SAILS AI programme.’

Pharmaceutical company on board

If the system could predict which molecule you need, an organic chemist would immediately be able to make a good molecule. That would save a lot of trial and error, which is ideal for the pharmaceutical industry. Pharmaceutical company Galapagos is working together with Van Westen. ‘We were awarded a grant for this from the Dutch Research Council (NWO), which Galapagos is supplementing. This is on the condition that the code that we develop becomes freely available, so to other businesses too. The advantage to Galapagos is that they will already be able to align the code with their working methods in the development phase.’

Together with other disciplines

The data driven drug discovery that Gerard van Westen is working on is much improved with the help of other disciplines. Together with computer scientist Michael Emmerich he therefore set up the Centre for Computational Life Sciences (CCLS)  in 2018, for researchers in the life sciences who combine their own discipline with computer science. ‘Research on zebrafish, for example, is also responsible for major breakthroughs that may be able to help us. And computer scientists look in a very different way at our data and make certain deliberations.’

A tangible project at the CCLS is a student exchange with Ukraine, led by Professor of Molecular Physiology Mario van der Stelt. ‘Students learn from this and come and visit us, and within a few weeks I’ll have the molecules delivered that I need for my own research.’

Text: Rianne Lindhout
Photo: Patricia Nauta

This website uses cookies.