© Chokniti Khongchum/Pexels
The AlphaFold software is a bit the new herobioinformatics and structural biology. This tool powered by AI was developed from 2018 by DeepMind (a subsidiary of Google) and it is particularly appreciated by researchers for its exceptional performance in terms of prediction of the three-dimensional structure of proteins. For example, he was able to identify hundreds of thousands of potentially psychedelic molecules (family of psychotropic drugs including very varied molecules, such as LSD, psilocybin or mescaline). For medicine, it is a springboard towards the development of new antidepressants.
In short, AlphaFold was a real game changer: its details are not limited to the boundaries of the academic field, but they have practical implications. There was a before and after Alphafold; Previously, determining protein structures was a very complex and time-consuming process. Certain methods that we used (cryo-electron microscopy, X-ray crystallography or nuclear magnetic resonance for example) could take months to produce usable results. Today, everything has changed.
A revolution in drug design
& #8220;AlphaFold is an absolute revolution” says Jens Carlsson, a computational chemist at Uppsala University in Sweden. In addition to having already revolutionized biology, AlphaFold now offers a colossal public databasecomprising approximately 350,000 protein structures. You need to retrieve the three-dimensional structure of hemoglobin ? Just type in the site's search bar and you're done!
For the pharmaceutical industry, these structures are essential to identify and improve promising drugs.
However, some specialists still temper their enthusiasm. This is the case, for example, of Brian Shoichet, pharmaceutical chemist at the University of California (San Francisco), who explains his point of view: “< em>there is a form of hype. Every time someone claims that this or that tool will revolutionize drug discovery, it calls for some caution“.
Even though AlphaFold is very accurate, its predictions are not always perfect, but ultimately, which tool really is ?
Putting AlphaFold to the test
To test the limits of the model, research was carried out by Shoichet and Bryan Roth, structural biologist at the University of North Carolina (Chapel Hill). The protein structures predicted by AlphaFold were therefore subjected to rigorous evaluation. They thus examined to what extent these structures were effective in the context of the search for new drugs. This analysis was carried out mainly by looking at how different chemical compounds interacted with these proteins.
One of the indicators checked was the success rate, i.e. the efficiency with which these interactions led to significant changes in protein activity . If the success rate is high, this means that a significant number of the compounds tested have succeeded in significantly modifying the activity of proteins.
< p>They compared the results predicted by AlphaFold, and those obtained by traditional experimental methods. To their surprise, the results were almost identical. Shoichet said: “c‘is a truly new result “. This means that AlphaFold is extremely effective.
Carlsson's team, in other unpublished work, found that AlphaFold's predicted structures for drug identification were effective and accurate in 60% of cases. case.
Limits and future potential
Even if these results are very promising overall, AlphaFold, like any model, has limits. Karen Akinsanya is president of research and development for therapeutics at Schrödinger, a drug software company based in New York. His opinion is clear about AlphaFold: “it’is not a panacea“ ;.
While some of the predicted structures are very useful for particular groups of drugs, this is not the case for all. Indeed, another study proved that AlphaFold could consider some of its predictions to be very precise in 10% of cases, but that ;#8217;they differed significantly from the structure found experimentally.
Nothing that significantly calls into question the potential of DeepMind's model. Indeed, according to Shoichet's estimates, a structure predicted by AlphaFold could accelerate a project in a third of cases. Although he admits that the predictions are not universally useful, he explains: “there are many models that we haven't even considered worth trying, because we thought they were of very poor quality“.
Isomorphic Labs is a subsidiary of DeepMind that specializes in drug discovery. The company hopes to capitalize on these gigantic advances by signing contracts with giants of the pharmaceutical industry like Eli Lilly and Company or Novartis.
AlphaFold is therefore a giant leap in the field of biology, and by extension medical research. Considering that this model is still in its infancy, the results it provides are exceptional. The potential it carries in the discovery of new therapeutic substances is undeniable, and it will certainly be a driver of medical progress in the years to come.
- AlphaFold is a protein structure prediction AI model that has the potential to revolutionize drug design.
- It has been released ;test by comparing it to classic experimental methods, and the results were really convincing.
- Even if it has its limits, DeepMind, l’ ;the company that developed it intends to take advantage of the power of its model by collaborating with large pharmaceutical groups.
📍 To not miss any news from Presse-citron, follow us on Google News and WhatsApp.