fbpx
  • DeepMind Used Deep Learning to Predict Protein Structures Associated with Corona Virus

    DeepMind researchers have released predictions of the protein structures associated with the currently spreading pathogen known as coronavirus or COVID-19.

    In an attempt to provide additional resources for the rapid development of tests for the novel pathogen, researchers employed the latest version of DeepMind’s AlphaFold system to predict the protein structures which are potentially connected to the SARS-CoV-2 virus. The system was developed recently at DeepMind and it uses deep learning to do “free modeling” of proteins or predicting protein structures when no structures of similar proteins are available.

    As a response to the virus outbreak, many scientific labs around the world contributed with their work on characterizing the pathogen and some of them have even shared computationally predicted structures of some of the viral proteins. DeepMind researchers mention that although their AlpaFold system is not perfect and they cannot be certain in the accuracy of predicted structures, the system provided accurate prediction for the SARS-CoV-2 spike protein, which was computationally discovered and shared in the so-called Protein Data Bank. In addition to that, protein structure predictions from AlphaFold were released for 6 more proteins: membrane protein, protein 3a, Nsp2, Nsp4, Nsp6, and Papain-like proteinase. All of the predicted structures of these proteins have not been experimentally verified by any lab in the world.

    The work was open-sourced and the predicted protein structures can be downloaded from here. More in detail about AlphaFold or the COVID-19 protein structure predictions can be read in the official blog post.