Why do we have to rack our brains to predict protein structure?

Did you know that there are hundreds of thousands of proteins in our human body, and all proteins are composed of only 20 kinds of amino acids. How do these amino acids form complex proteins? How do scientists explore the functions of different proteins? I heard that scientists have recently applied the most advanced artificial intelligence technology to protein prediction. What is going on?

Today we are going to talk about proteins that are closely related to all functions of our human body, and the way to predict protein structure.

Proteins are molecules with very diverse structures and functions, and the reason why their functions are so diverse is due to the different arrangements and combinations of amino acids in the molecules. Amino acids are the smallest molecular units in proteins. Two amino acids holding hands can be dehydrated and condensed into a dipeptide. Similarly, multiple amino acids can be condensed into a peptide chain. The peptide chain condensed in a certain order is called the primary structure of protein.

So the question is, how are the different amino acids arranged on a peptide chain? This depends on genes! Genes are functional fragments of DNA. They produce functional proteins through transcription, translation and expression, thereby controlling the life activities of cells and then the entire organism.

How are these complex proteins formed? When different amino acids are arranged into peptide chains according to the instructions of genes, they will curl or fold to form a secondary structure. At this time, the arrangement of amino acids is either like a spiral surface or a folded sheet.

When the secondary structure folds again in a variety of ways to form a spherical or fibrous three-dimensional structure, the tertiary structure is formed. This specific folding method allows different peptide chains to have different shapes to perform different functions.

Do you think this is the end? In fact, if a protein has more than one peptide chain, the multiple peptide chains will further curl and fold to form a quaternary structure.

Therefore, the way a protein folds - that is, its structure - determines its specific function and also determines the function of our organism. Therefore, predicting unknown protein structures is very important and can help us unlock the unsolved mysteries of the functions of living organisms.

This is why scientists have been constantly exploring the structure of proteins. In the past few decades, we first saw the double helix structure of DNA from X-ray crystal diffraction patterns. In the following 40 years, scientists relied on this method to resolve the structure of most proteins.

However, it is very difficult to crystallize some proteins. For this reason, in the 1970s, scientists in Cambridge, UK, developed a method for analyzing protein structure using cryo-electron microscopy. Since then, we have unveiled the mystery of some macromolecules and membrane protein structures that are difficult to crystallize. Some basic receptor proteins and transcription complexes, the most basic molecular structures in life activities, have been analyzed. This field has also produced several Nobel Prizes, so you can imagine how important it is.

Afterwards, with the vigorous development of Internet technology, the capabilities of computer algorithms have advanced by leaps and bounds. DeepMind, which became very popular in the past two years, became famous in the man-machine battle and has now moved to another more complex field - predicting the structure of life macromolecules. DeepMind has developed AlphaFold, which can directly realize the magical prediction of sequence to structure. Even the extremely challenging prediction of protein interactions can be cleverly solved by combining AlphaFold with other algorithms.

In addition to Deepmind, Americans are not to be outdone. David Baker's team at the University of Washington, Seattle, used the RoseTTAFold[rəuˈzetə] software to create a powerful protein structure analysis tool that can complete missing parts of the protein structure when the gene sequence is unclear.

When it comes to this popular technology track, our country is naturally not willing to be outdone and has made some gratifying progress, but it is only secondary development based on DeepMind's open source data. There is no real original technology, and it will still take some time to achieve a breakthrough from 0 to 1.

I believe that in the future, more advanced and innovative technologies and methods will make protein prediction technology more powerful.

This article is a work supported by Science Popularization China Starry Sky Project

Team Name: Deep Science

Review: Tao Ning

Produced by: China Association for Science and Technology Department of Science Popularization

Producer: China Science and Technology Press Co., Ltd., Beijing Zhongke Xinghe Culture Media Co., Ltd.