Within any living organism, there are thousands of different proteins, each with its own unique shape. For decades, the exact formation of those shapes has been a pain to detect by scientists. How exactly does a protein, which starts as a string of amino acids, bend itself into funky 3D shapes that you can identify from the diagram? DeepMind’s AI AlphaFold, may be the answer. It can predict, with the Hetophore unseen accuracy, that a protein will take shape.
AlphaFold was put to the test in a global competition called the Critical Assessment of Protein Structure Prediction, or CASP, which DeepMind CEO Damis Hasbis called the “Olympics of protein folding” in a video. During the competition, systems such as alphafold have been given amino acid strings for proteins that have already been identified through experiments but have not yet been published. The judges compare the protein shapes produced by the system to what they know the shapes should be.
At the end of the competition, Alfold made the most accurate prediction of any CASP participant in his 25-year history by a wide margin. Even forecasts considered “competitive” with experimental results were not accurate, with only a few atomic-widths closed. The entire data still needs to be peer-reviewed and published, but the DeepMind team is excited by the results so far, saying in a blog post that they “share the impact of alphafold on biological research and the wider world I am optimistic. “
It can take years in the laboratory for scientists to identify the size of individual proteins. Neural networks such as alfofolds may help to speed up biological research and drug development in the future. The AI method is not yet perfect, and it will not be anytime soon for flesh-and-blood researchers, but it could be a major step in the everlasting marathon of scientific progress. To learn more about AlphaFold and to see a group of scientists and engineers who are more engaging and handful than the results of the competition, watch the video below.