With advances in artificial intelligence (AI), researchers can now create molecules very quickly. Previously, it took several months to design a molecule. Today, a team of scientists from the University of Washington in the United States has created new proteins in seconds. The results were published in Science.
Artificial intelligence at the service of biochemistry
Last July, the artificial intelligence software AlphaFold from the company DeepMind belonging to the giant Google succeeded in predicting the structure of all the proteins known to date by scientists. AlphaFold software is an artificial intelligence tool designed as a deep learning system. Also called “deep learning”, deep learning is a method of automatic learning. It is based on the use of layers of artificial neural networks and series of complex propositional calculations.
Alpha Fold manages to predict the protein structure by their amino acid sequence. He studies the description of the sequence of amino acids that make up a protein. The quality achieved by version 2 of the software has made it possible to achieve exceptional levels of precision. It is free software, accessible free of charge. It gives access to a databases grouping together the structure of a phenomenal number of proteins, including 20,000 human proteins.
Last June, the South Korean authorities authorized the marketing of a vaccine intended to fight against Covid-19. This vaccine is based on the use of fragments of the spike protein of the SARS-CoV-2 virus. But instead of “extracting” them from the virus itself, these fragments are derived from recombinant proteins and then assembled in the form of nanoparticles to resemble the structure of the virus. It was thanks to the work of researchers on nanoparticles, about ten years ago, that this was possible.
Today, thanks to artificial intelligence, it is possible to design such nanoparticles in just a few seconds!
Read also: Spike protein of the Covid-19 virus: software predicts its interactions with cell membranes
Proteins that adopt the correct conformation
Researchers manage to create new molecules, but which are of no use. They want to be able to create proteins whose structure could be used, for example, to eliminate pollutants or to treat diseases.
In the early 1990s, researchers developed software called Rosetta. It made it possible to manufacture proteins in several steps. The software made it possible, among other things, to deduce the amino acid sequence corresponding to a given protein. Unfortunately in many cases, the proteins did not adopt the correct conformation, that is to say their arrangement in 3D in space. An additional step had to be introduced to modify the sequence of the protein so that it adopts the correct structure. This step required enormous computing power.
Thanks to artificial intelligence and the latest software AlphaFold, this step has become almost instantaneous. In one approach called ‘hallucination’, scientists feed the amino acid sequence into software. This will then predict the closest structure allowing this assembly of amino acids to resemble a protein. The scientists found that when making these proteins in the lab, more than 1/5 of them matched the structure predicted by the software.
The scientists discovered that the software could also predict and model the assemblies of several interacting proteins. So they decided to use it to create proteins. These could self-assemble to form nanoparticles of different shapes and sizes. Unfortunately when they decided to go hands-on and create these molecules in their lab, the synthesized molecules did not fit together at all.
Read also: deep learning
A revolution in the design of new molecules
To address this conformation problem, the scientists used a new learning tool called ProteinMPNN that can correct the folding problems of the proteins created by AlphaFold by modifying the sequences. Thanks to this new tool, they succeeded in creating protein nanoparticles with complex symmetries.
These deep learning tools are a revolution in the field of protein engineering, a relatively young discipline. The latest developments make it possible to design a protein using dedicated software and to launch the design immediately with a success rate of 10%. By combining several neural networks like AlphaFold and ProteinMPNN, it is possible to increase this rate.
There are currently many tools for designing proteins by artificial intelligence. Among them, a large number tackle the problem of protein folding such as proteinMPNN software. However, faced with this large number of tools, it is not always easy to find the right tool.
Google’s DeepMind ventured into drug research last year by founding Isomorphic Labs in London. Its goal is to apply artificial intelligence tools like AlphaFold to the discovery of new drugs. This new revolution in rapid protein design will undoubtedly find its application within this company.
Read also: An artificial intelligence, a million times faster than the human brain!
We would love to say thanks to the writer of this post for this outstanding material
Artificial intelligence and very fast design of molecules
Discover our social media profiles as well as other related pageshttps://www.ai-magazine.com/related-pages/