Artificial intelligence has learned to create new proteins

(ORDO NEWS) — For the first time, scientists have used artificial intelligence to generate completely new proteins. By combining machine learning algorithms, the researchers were able to create complex protein structures with bizarre shapes that could still function.

Over the past two years, artificial intelligence (AI) has revolutionized protein structure prediction. Now a similar breakthrough has occurred in the field of protein design.

Scientists from the University of Washington (USA) have shown that machine learning can be used to quickly and accurately create new protein molecules.

The results of their research, published in the journal Science, will lead to the development of new vaccines, drugs, biomaterials and carbon capture devices.

Proteins are an integral part of any living organism, they are involved in almost all cellular processes. A protein is made up of sequences of amino acids that define its three-dimensional shape, which is essential for its function.

Not so long ago, the machine learning algorithms AlphaFold and RoseTTAFold learned to predict the structures of natural proteins in detail based on amino acid sequences alone.

However, all existing proteins make up less than one percent of all theoretically possible proteins, which can also have various unique and useful functions.

Now scientists have shown that AI can generate new protein structures in two ways. The first is called hallucination: it resembles DALL-E or another generative neural network that generates an output based on simple clues. The second, inpainting, is similar to the autocomplete feature in modern search engines.

To speed up the process of creating proteins, scientists developed a new ProteinMPNN algorithm for generating amino acid sequences, which was 200 times faster than the best of previous analogues.

In addition, the researchers used the AlphaFold program to independently evaluate whether the amino acid sequences that ProteinMPNN created could fold into desired shapes.

The combination of these machine learning tools has created new proteins that have functioned successfully in the lab.

Scientists managed to build some very complex protein structures. For example, nanosized rings, which, according to the creators, can become parts of non-standard nanomechanisms.


Contact us: [email protected]

Our Standards, Terms of Use: Standard Terms And Conditions.