Breakthrough in protein structure prediction
Biochemists have dreamed of being able to read the exact three-dimensional structure of a protein from the amino acid sequence for decades. With the help of self-learning algorithms, they have now come much closer to this goal.
Proteins are responsible for almost all chemical and mechanical processes in cells. They are made up of amino acid residues that form long chains and spontaneously fold into three-dimensional structures. The sequence of amino acids determines a protein's precise shape and range of motion, which in turn determines its function. Biologists have experimentally examined thousands of these structures over decades, but this is extremely time-consuming. So some of them turned to other methods to find out what shape a protein might take. Andrew Senior of technology company DeepMind and his colleagues have now developed the AlphaFold algorithm that uses AI to tackle the problem - and has already led to major advances.
Because of the numerous possible three-dimensional structures that a protein can form, no simple folding rules can be derived. This is what makes it so difficult to predict what an amino acid sequence will look like as a protein. Ultimately, quantum mechanics determines the exact protein structure. If the exact energy could be calculated for every possible shape, one would only have to select the energetically most favorable conformation. However, each protein has an enormous number of different folding possibilities, which actually rules out such a quantum mechanical approach …
