[ad_1]
A’artificial intelligence she managed to “dream” and design different molecules, the design of which had required human intelligence many months of work, in a matter of seconds. She tells it, in two articles published in the magazine Science (here And here), a team of biochemists from University of Washington in Seattle, coordinated by David Baker: the news (together with that dating back to about two months ago, when the experts of DeepMind have announced that the AlphaFold managed to predict the structure of almost all known proteins, just awarded the prestigious Breakthrough Prize 2023) it makes the beginning of one even more concrete and close new era in pharmacologyan era in which artificial intelligence will completely change the times and ways of designing and developing therapeutic molecules.
Why “designing” proteins is so difficult (and so important)
It all depends on the fact that their structure is very, very complex. There three-dimensional structure of proteins – compounds that are among the constituent elements of living beings – is closely linked to the function they perform: although the individual units of which they are made, the amino acidsthere are only 20, arranged in precise sequences, in nature there are millions of different proteins, each of which has different features and functions. This is possible because the amino acids are arranged in space in a different way according to their chemical characteristics; therefore, each amino acid sequence of which a protein is made up will correspond to a different structure. There are many three-dimensional combinations, and only a part of them (a very high number) corresponds to actually “possible” proteins: understanding the three-dimensional configuration of a protein is fundamental in research, because provides information about its position and how to change, lock or adjust it; over the years, the study of the three-dimensional structure of proteins has in fact demonstrated its usefulness in many areas of the life sciences, including the discovery and synthesis of new drugs.
It must also be taken into account that while the amino acid sequence is quite simple to identifyit is not so obvious to derive the three-dimensional structure from it: traditionally – that is, before the advent of artificial intelligence – we resort to experimental techniques which however present various obstacles in terms of complexity and implementation times. Obtaining a prediction of the protein structure with a “classic” computational approach was previously possible with bioinformatics technologies, but only for limited parts of the molecule. What artificial intelligence has triumphed, completely changing the cards on the table, was to find a way to obtain, starting from an amino acid sequence, a reliable and high-resolution prediction of the structure of a protein in its entirety. To put it in the words of the experts, it is as if the artificial intelligence “dreamed” the structure of the protein once the sequence of its components is known. A sort of “hallucination”, in short.
The turning point
“Since the arrival of AlphaFold – has explained to Nature Noelia Ferruzcomputational biologist atUniversity of Gironain Spain – there has been a radical change in the way we work on protein design. It’s an extremely exciting time “. And it is in this context that the work of Baker and colleagues fits, a work begun over thirty years ago, when scientists began to develop Rosette, a software that dealt with the prediction of the protein structure by dividing the process into several steps: the researchers tried to imagine the structure of a new protein, often putting together pieces of other proteins, and the software tried to infer the sequence of amino acids corresponding to this structure. The problem is that this process, in addition to being very long and expensive from a computational point of view (“You had to have – he says, still a Nature, Sergey Ovchinnovevolutionary biologist at Harvard University and former member of Baker’s research group – about 10 thousand computers in action simultaneously for several weeks “), was also imprecise: the “drafts” created by the software, once reproduced in the laboratory, rarely took the desired shape, and it was therefore often necessary to further change the structure of the protein several times before arriving at the result. A process of “trial and error”, in short, long and complicated.
AlphaFold’s artificial intelligence and the like made these steps virtually instant. In the new approach of Baker and colleagues, in particular, called precisely hallucination, scientists insert random sequences of amino acids into a network that predicts their structure, altering it more and more until it becomes similar to a “possible” protein, according to the software itself. With this system, Baker’s team created about 100 “hallucinated” proteins in the laboratory last year, about a fifth of which closely resembled the expected form. And then came today’s work, in which scientists presented the results of a new instrument, ProteinMPNN, which addresses the so-called “reverse folding” problem by specifying an amino acid sequence that corresponds to a particular protein structure; and in this case too the results were extremely encouraging. The Baker and AlphaFold groups, of course, are not the only ones working on the theme: in one review recently posted on the server bioRxivFerruz and colleagues counted over forty artificial intelligence algorithms dedicated to protein design, all developed in recent years.
To understand the importance of these technologies it is good to remember some of the results obtained and expected: last year a team of scientists from University of Oxfordin the UK, used artificial intelligence to study a protein that it would represent one of the most promising candidates for the development of a vaccine against malaria, trying in particular to understand in which point of the structure the most effective antibodies could bind to block the transmission of the parasite. Among the other areas in which this technology has been used there is also the basic research in biology, but not only: AlphaFold has also been used in some studies onplastic pollutionon the Parkinson’s diseaseon the bee healthon the ice formationon neglected diseases and onhuman evolution.
.
[ad_2]
Source link
