‘My algorithm can help researchers to understand processes in the body’

If we want to understand how we get ill, we need to understand how our body works and in what way molecules like proteins and our DNA influence certain processes. Molecular dynamics simulations are a good way to study what is going on in our body, but with all these different and complex molecules the simulations can get pretty complicated. During his graduation project in the Molecular Dynamics Simulation Group of the University of Groningen, Stylianos Mavrianos tried to predict the end state of the protein that is being simulated using deep learning algorithms.

Molecular dynamic simulations are not for the faint-hearted. If you want to study proteins or processes in the human body, there are many different factors that can influence the components. ‘The field of molecular dynamics simulations combines chemistry, biology and physics’, Stylianos says. ‘That scared me a bit, but it was also a nice challenge because these simulations can really help researchers to understand how we get ill and find a cure.’

To achieve this, Stylianos used a deep learning approach on simulated data of the so-called EGFR protein, a well-studied protein that plays a role in the proliferation of cells. Mutation on this protein can often lead to cancer. Stylianos: ‘Because EGFR is a well-known system, I could test my algorithm and see if my predictions were the same as researchers saw in experimental data. At first it was a bit off, but with optimisation I could improve the algorithm and make it more accurate.’ Despite his promising results, the algorithm has only been tested on one system. In order to validate the algorithm more testing is needed, Stylianos says: ‘Preferably with different also well-studied systems, to see if the results are congruent with experimental data. If that is the case, then the algorithm can be used on lesser known systems and give us insight on how these proteins work.’

As a result, Stylianos developed a python package that can generally be used by everyone who works on molecular dynamic simulations. ‘In principle you can just put in your protein trajectories data and the algorithm will tell you which amino acids of these proteins are most important for the state that the protein is in.’ He is happy that it worked out in the end. ‘I think this tool can be used as a stepping stone to help researchers make advances in understanding and treating diseases.’

Now that he has graduated, the research group will continue the work on this algorithm while Stylianos looks for a job. ‘I am very interested in environmental problems and especially plastic pollution, and data science can really play an important role in this field. For instance, I want to build a predictive algorithm that models how plastics spread through the environment which would then help us in finding a way  to stop this. I hope to find a job where I can use my knowledge to help with these kinds of problems.’