Alphabets DeepMind has achieved a new scientific breakthrough. Company researchers have developed an AI system that can simulate the form of proteins. The system has been in development for two years and the fact that it is proving to be working can bring about a whole new era in the world of medicine.

According to The Guardian, the researchers at DeepMind have decided to announce the project, after the software turned out to be better at 97 other algorithms during a CASP simulation competition of the American National Institute of General Medical Services. AlphaFold, as Artificial Intelligence is called, was able to predict the 3D shapes of proteins.

What are proteins

Proteins are molecules that perform a range of important functions in organisms. They give cells their structure, enable metabolism, DNA replication and react to stimulation. Proteins also transport molecules from one location to another. The human body can produce huge numbers of different proteins. How much is not certain; estimates range from a few tens of thousands to many billions.

Each protein is a sequence of amino acids, of which there are twenty. A protein can bend between all amino acids in all kinds of forms. A protein with hundreds of amino acids can therefore take on almost countless different structures. CASP is a global company that has taken place every two years since 1994. The abbreviation stands for Critical Assessment of protein Structure Prediction, in which scientists make an attempt to predict the form of proteins.

Sizeable performance

The shape of a protein depends, among other things, on the number of amino acids it contains. Its shape also determines the role that a protein plays in the human body. A heart cell, for example, has proteins that are shaped in such a way that adrenaline sticks to it, which can be used to increase the heart rate. An antibody is also formed in such a way that it binds malignant cells to itself and can carry them away. Sun bit all the functions in the human body – from muscle movements to changing nutrition to energy – can be traced back to the shape and movement of proteins.

Predicting this is extremely complicated. But this year AlphaFold correctly predicted the form of 25 out of 43 proteins. By way of comparison, the number two on the list could correctly predict three of the 43 proteins. So it’s a great achievement: This is a key moment for us, says co-founder and CEO Demis Hassabis of DeepMind. This is our first major investment of people and resources in a fundamental, extremely important, real scientific problem.

Great expectations

AlphaFold’s system works with a neural network that is trained on thousands of already known proteins. That network was trained until it could predict the 3D form of amino acids. If AlphaFold now has to predict the shape of a new protein, it predicts the distance between amino acids and the angle of the chemicals that bind them together. The software then searches for the most energy-efficient composition. Whereas AlphaFold initially took more than a day to predict the shape of a protein, it now only takes a few hours.

The expectation of experts is that this software should eventually be able to help with the development of new medicines and new treatment methods.

This news article was automatically translated from Dutch to give a head start. All news articles after September 1, 2019 are written in native English and NOT translated. All our background stories are written in native English as well. For more information read our launch article.