Google AIdeveloped by the company DeepMind, has been studying proteins for years as a first step toward interfering with many biological processes. The first was AlphaFold, an algorithm that can predict how proteins fold. Now it’s the turn of AlphaProteonew AI that can create proteins that interact with other proteins, opening or closing doors to intervene in a very wide range of phenomena with different medical applications.
Indeed, biological processes are controlled mainly interaction between proteins. Some click together like a key in a lock, triggering a cascade of reactions that lead to the target. If necessary, proteins can be used to block the lock or enhance its effect. But no matter how much we understand what happens after the interaction, designing a key from scratch is difficult.
That’s why DeepMind scientists decided to take Google’s AI one step further. They developed an algorithm that can Rapidly create proteins with strong binding capacity. At this point, almost everything is done. in silica or, equivalently, with computer models. However, some laboratory experiments have already been conducted, and the results confirm what scientists have already seen on their screens.
Importance binders squirrel
For a protein to function properly, it must have correct folding. It could be perfectly formed with all its amino acids, but not fold correctly, losing all its functionality. It could also fold slightly differently, giving it a new function. All of this was analyzed using AlphaFold. It was very interesting and useful, but we needed to go even further. We needed to have what is called binders proteins.
These are small proteins that bind specifically to other target proteins so that their functions can be visualized or altered. Initially, binders were most often used in research. antibodies. These are small proteins that bind to other proteins, usually to mark them so that the rest of the immune system will attack them. For example, we may have antibodies that can recognize specific proteins on SARS-CoV 2, in case we get infected by it.
Similarly, there are antibodies to many other proteins, even if they are not pathogens. The problem is that it was not always possible to obtain an antibody that concrete and stable for a specific function. So over time, others binders.
What does Google AI do?
Google’s AI has been trained with over 100 million data to give rise to AlphaProteo. All of these data relate to protein binding structures with a known function.
By analyzing all this data, Google’s AI was able to quickly develop new bindings, taking into account three aspects. On the one hand, protein structure which they must join. On the other hand, different attachment points binders what’s in it. And finally, everything is different folds.
Once the protein is produced, it is tested using computer modeling to see if it can bind tightly to the target protein. And the truth is, the results exceeded the researchers’ expectations.
Milestones Achieved by Google AI
binders obtained with the help of Google AI, can join from 3 to 300 times stronger to the target protein than others obtained by other methods. In addition, milestones were achieved that no other algorithm had previously achieved.
For example, for the first time A. binder which effectively and firmly binds VEFR-Aprotein involved both in cancer and in complications of diabetesSeveral binders have also been obtained that are capable of binding to domains of proteins associated with the infectivity of viruses and bacteria.
Good results in the laboratory too
Results in silica left no room for doubt about the effectiveness of Google’s artificial intelligence. But it was also necessary to test it in the lab. Some of binders AlphaProteo’s developments were synthesized in the laboratory and real tests were carried out, the results of which turned out to be almost the same as those of computer models.
So DeepMind has taken another giant step. AI can bring us many problems, but also many benefits. What Google’s AI achieves certainly falls into the latter group.
Source: Hiper Textual
