Exploring the structure of viral proteins can help develop targeted drugs.

The prediction of AlphaFold's protein structure of the new coronavirus is a structural reconstruction that is separatefrom the experiment.
the accuracy of the predictions, peer review and validation of actual clinical treatment.
proteins are complex biological macromolecules necessary to sustain life, and almost all functions in the body, such as muscle contraction, breathing, or converting food into energy, are closely related to the interaction of proteins.
obtainath of the protein's three-dimensional structure helps scientists understand its role in the human body and design the corresponding drugs.
, deepMind, an artificial intelligence company, announced that it had used AlphaFold to predict the three-dimensional structure of six proteins encoded by the new coronavirus gene, including membrane proteins and non-structural proteins, and that it was open for download.
learned that the structure of viral proteins could help develop targeted drug viruses made up of nucleic acids and proteins, which are encoded by the virus genome. There are two
viral proteins, one is structural proteins, which can form a mature form of infectious virus particles, helping the virus infect cells, such as shell proteins, membrane glycoproteins and enzymes present in viral particles, and the other is non-structural proteins, which help the virus replicate in host cells, gene expression, expand the "territory" in the human body.
as early as January 10, China announced the new crown virus genome-wide sequence.
but just knowing the genome sequence doesn't fully understand how proteins work.
protein consists of 20 amino acids, each consisting of dozens to thousands of amino acids. The linear sequence of
parts of amino acids will form a spiral or folding secondary structure, and further orderly combination of three-dimensional structure, this three-dimensional structure determines how the protein in the human body to function.
" Xiao Yibi, a professor at the School of Pharmacy of China Pharmaceutical University, told Science and Technology Daily, for example, if the human body's virus receptor is a lock, the virus's stingy glycoprotein is the key, if these keys can be inserted into the human virus receptor protein, will infect the cells, scientists need to do is to find out what the three-dimensional structure in the key, the key and lock relationship is what, and then prevent the key to unlock, that is, to prevent the virus from infecting cells.
know how a protein works, you know how to inhibit viral activity, and if a protein is found to be the key protein to invade the host cell, you can do a drug design for a protein or an area of the protein.
, said Dong Xianchi, a professor at the School of Life Sciences at Nanjing University.
the predictions even if the exact experimental process is unavoidable in the DeepMind team's view, can determine the three-dimensional structure of proteins based on amino acid sequences.
they used two methods to build predictive models by predicting the distance between each pair of amino acids in proteins and the angle between the chemical bonds that connect them.
" the first step is to train neural networks to predict the distance or angle of each pair of amino acids in a protein, and then continuously combine these probabilities to improve the accuracy of protein structure prediction, and the second step is to optimize the score through gradient decline.
they predicted the entire protein chain, not the protein 'fragments' before the protein structure was assembled, thus reducing the complexity of the entire prediction process to some extent.
" Peng Shaoliang, deputy director and professor of Hunan University's Supercomputing Center, told Science and Technology Daily that AlphaFold modeled the morphological structure of proteins from scratch rather than using the resolved proteinas as a template, which means that large amounts of computation are required.
according to Wang Xiao, associate professor of bioinformatics at Tsinghua University's Department of Automation, there are about 30,000 known protein structures in the current international protein database (PDB), which can support protein structure prediction using protein sequences that are similar to the target sequence.
in addition to the depth of artificial intelligence learning, scientists want to acquire protein structures, mostly from MRI, cryoscopy and X-ray diffraction technology.
" all three methods rely on large-scale facilities, instruments, experimental means to obtain the protein structure, colloquially speaking, is to give protein multi-angle photographs, and then according to the massive two-dimensional photos reconstruction of three-dimensional structure, the results are objective and accurate, but the experimental cycle is relatively long, usually required for several months, the experimental threshold and experimental cost is high, the experimental difficulty is not small.
," Peng said.
alphaFold's prediction of the protein structure of the new coronavirus is a structural reconstruction that is separatefrom the experiment.
the accuracy of the predictions, peer review and validation of actual clinical treatment.
, however, DeepMind notes, "the model will point out which parts of the structure are more likely to be correct, and while these unstudied proteins are not the focus of current treatments, they may increase researchers' understanding of the new coronavirus."
and for AlphaFold's prediction, Peng Shaoliang believes that if the prediction results are accurate, there are many computational and analytical processes such as molecular docking, molecular dynamics simulation, and the validation of animal experiments and human clinical trials.
" calculations can be accelerated, but the experimental process is unavoidable, and ultimately everything is aimed at making clinically available drugs and vaccines.
"Source: Science and Technology Daily Jin Feng.