Artificial intelligence predicts protein structures
| Reading time: 3 minutes
Proteins control various biological processes in the human body. Artificial intelligence can now determine its three-dimensional structure. The development of individualized drugs will benefit from this.
PReds, also called proteins, are all-rounders in the human organism. They not only give cells their individual structure, but they control all kinds of biochemical processes. Muscles, heart, brain, skin and hair even consist mainly of proteins. Researchers have now succeeded in predicting the exact structure of protein molecules with the help of artificial intelligence with an accuracy that was previously impossible. This opens up new perspectives for pharmaceutical research.
The entirety of all proteins in a human body is called a proteome. It varies a little from person to person, because the instructions for building the proteins are in the genetic make-up, the genome. And this genome is individually different. Different eye and hair colors are a consequence. If, however, the blueprint for a vital protein is missing in the genome, it is the cause of a hereditary disease. Due to the central importance of the proteome for functionality and the tendency towards certain diseases, it is sometimes referred to as the “second genetic code”.
Proteins are intricate balls
Proteins are macromolecules that consist of amino acids strung together. These chains, however, tangle up to form extremely complex spatial structures and are by no means linear threads. The structure of proteins is not stored in genes, but ultimately results from the laws of quantum physics. The protein folds in such a way that the structure allows the lowest possible energy. However, there are many alternative options with quite similar energy values.
Which biological properties a protein has and which tasks it can therefore perform in the organism, depends largely on its three-dimensional structure. This can be determined with the help of a technically very complex X-ray crystal structure analysis. To do this, however, you first have to create a crystal from the proteins in question. The structure of less than one percent of all proteins in the human body is known so far.
So far, it has not been possible to calculate, based on a known sequence of amino acids, how the protein in question folds three-dimensionally and what properties it will therefore have. The solution to this “protein folding problem” is particularly important for the development of drugs in which tailor-made proteins are supposed to intervene therapeutically in the metabolism.
Artificial intelligence is revolutionizing biology
Now the British company DeepMind, a subsidiary of the Google holding company Alphabet, has reported a breakthrough in the precise prediction of protein structures. Your AlphaFold program uses artificial intelligence (AI), so-called deep learning algorithms. This could revolutionize not only pharmacy, but also synthetic biology. Because on the basis of already known protein structures, completely new molecular machines can be created through targeted protein design.
“With the help of machine learning, the AlphaFold team succeeded in extracting the rules of protein folding from a large number of known protein structures so well that it was able to predict 70 out of 100 protein structures very precisely in a blind test competition,” says Professor Helmut Grubmüller from Göttinger Max Planck Institute for Biophysical Chemistry.
“Wow, this is a breakthrough!” Comments Jan Kosinski, group leader at the European Molecular Biology Laboratory (EMBL), “the accuracy and success rate of AlphaFold is unparalleled. I can hardly wait to use this method for my proteins. ”But Kosinski points out that there are still other challenges:“ Proteins do not usually take on a single fold, but change in response to the environment in which Binding to other proteins, when performing enzymatic reactions, or when binding drugs or therapeutic antibodies. It seems that AlphaFold cannot yet predict these changes. But deep learning can also be applied to such problems. “
With regard to the medical applications of AlphaFold, Professor Gunnar Schröder from Forschungszentrum Jülich is optimistic. The head of the Computational Structural Biology research group says: “The rapidly growing field of personalized medicine could even become personalized molecular medicine, in which we will in future tailor active ingredients and therapy to the personal protein structures of a single patient.”