Google's AlphaFold deciphers "God's code" to solve problems that have plagued mankind for 50 years

To the general public, Google AlphaFold seems familiar, and this acquaintance comes from its twin brother, AlphaGo, which defeated the Go master

In 2020, at the 14th International Protein Structure Prediction Competition (Critical Assessment of Protein Structure Prediction, CASP), AlphaFold2 successfully predicted the three-dimensional structure of the basic molecule of life-protein based on the gene sequence, and achieved a median score of 92.

Research results (Source: Nature)

Where is AlphaFold2 so powerful?

Technological Breakthroughs That Disrupt Medicine

Protein is the executor of all functions in the organism.

Proteins are generally polypeptide chains composed of tens to hundreds of amino acids dehydrated and condensed.

Central dogma (Image source: [4])

Proteins are like small and delicate biological machines, and the structure of the machine determines its function, so revealing the structure of the protein can help us understand the function of the protein

At present, the main way to carry out protein structure research is to analyze protein structure with the help of experimental instruments, such as X-ray diffraction, nuclear magnetic resonance technology, cryo-electron microscopy technology and intelligent computing prediction

Table 1.

Data source: Journal of Medicine and Philosophy [5]; Tabulation: BioQuest editorial team

Take the new crown vaccine as an example: its research and development is based on the fact that we have drawn a structural map of the cellular spike protein on the surface of the virus that is used to invade the human body

From a practical point of view, the success of AlphaFold2 has helped us overcome many major diseases

AlphaFold2 Multi-Field Technological Revolution

Tailor-made new proteins

AlphaFold2's fast and accurate resolution of protein structures allows scientists to create proteins from scratch, i.

In late January 2020, DeepMind scientists used AlphaFold2 to map the protein structure of the SARS-COV-2 virus — which was later experimentally confirmed to be accurate

In projects with limited research funding, AlphaFold2's free resources can be very helpful

DNDi and a team of infectious disease researchers from the University of Washington, Dundee and GlaxoSmithKline have discovered a molecule that binds to a protein on Trypanosoma cruzi, the parasite that causes Chagas disease disease

The Enzyme Innovation Centre at the University of Portsmouth, UK, applied AlphaFold2 to design proteins for processing single-use plastics, which do not exist in nature

The application of AlphaFold2 in the pharmaceutical field is controversial

Algorithms still need to be improved

The basic principle of AI structure prediction is to use proteins with known three-dimensional structures as data sets for a large number of training, input the amino acid sequence of a protein to calculate its three-dimensional structure, and compare it with the experimental structure of the protein, so as to strengthen machine deep learning capabilities and Predictive power for unknown protein structures

That is to say, the amazing results achieved by AlphaFold2 this time are completely inseparable from the proteins whose structures have been figured out by experimental biologists as templates for their comparison or learning

The so-called protein structure prediction is only an intermediate stage of structural biology research.
Whether it is experimental structure analysis or AI structure prediction, the ultimate goal is to understand the mechanism of life
.
As Professor Yan Ning of Princeton University said: "The subject of structural biology is biology, understanding life, and making biological discoveries
.
" From subatomic to atomic to molecular level is still a physical and chemical process, why is the molecular level? The proteins and their constituents are chemically active and capable of reproduction and evolution? That is, what and how has the protein changed at the molecular level? Answering these questions comes close to understanding life
.

But as mentioned above, understanding protein function requires understanding its structure, and structural analysis is currently far beyond human cognition
.
It is precisely because of this that many experimental biologists put a lot of time and energy into the experimental analysis of protein structures, and it is precisely because of the slow and difficult development of this process that people are prompted to find a new way to carry out structure prediction with the help of computational science
.
In this sense, experimental analysis or algorithm prediction is a tool and means for carrying out protein function research and understanding the mechanism of life, but it cannot be said that because this is still very difficult and is the main work of structural biology at present, as a tool and means The structural elucidation or prediction of it becomes the purpose of biology
.
In short, AI has made amazing achievements in the field of protein structure prediction, but this does not mean that AI clearly tells us the process and principle of protein folding
.

While AlphaFold2 was an eye-opener for the judges at CASP14 and has been used in a variety of research fields, this is just the beginning of such computational techniques
.
There may be 10,300 conformations of a protein, but AlphaFold2 still cannot answer how the protein spontaneously folds into the correct shape in an instant
.
AlphaFold2 currently represents the gold standard for AI protein prediction, but this benchmark will continue to improve as the technology develops and evolves
.

Accurate prediction of protein structure, which is beneficial in the design of therapeutics, allows researchers to visualize the shape of a target protein
.
However, the current limitations of AlphaFold2 mean that the field of drug design has yet to change significantly
.
Predicting the shape of larger multidomain protein complexes and knowing the positions of all amino acid side chains is important for designing drug molecules: these are areas where AlphaFold2 is currently difficult to predict
.
A recent paper also highlighted that while the structural data predicted by AlphaFold2 may shorten the time to early research, it is unlikely to drastically reduce the time it takes for a new drug to get from the lab to the patient
.

References:

[1]Jumper J, Evans R, Pritzel A, et al.
Highly accurate protein structure prediction with AlphaFold.
Nature.
2021 Aug; 596(7873):583-589.
doi: 10.
1038/s41586-021-03819-2.
Epub 2021 Jul 15.
PMID: 34265844; PMCID: PMC8371605.

[2]GitHub - deepmind/alphafold: Open source code for AlphaFold.

[3]AlphaFold Protein Structure Database.
alphafold.
ebi.
ac.
uk.

[4] http://sciencewithmsjones.
weebly.
com/living-environment/central-dogma-of-genetics?fbclid=IwAR13Gz372IrbOBKoNi3BWheUb9gwbrCgW30oApkw1lEwl19EHLMkJ75lAzI].

[5] Zhao Yunbo
.
Can AI predictions replace scientific experiments? Medicine and Philosophy, 2021
.
DOI: 10.
12014/j.
issn.
1002-0772.
2021.
06.
04