Proteins – in notes and vice versa

The composition of proteins was translated into musical melodies. And then they taught the neural network to compose protein plays

This became possible after the amino acid vibrations were transposed to a human-audible frequency.

Maxim Abdulaev, "The Attic"

Scientists from the Massachusetts Institute of Technology transposed the vibrations of all amino acids so that they could be heard by the human ear. Then they translated the proteins into "amino acid melodies" and, having trained the neural network on these musical tracks, forced it to compose similar music. It is structurally similar to the melodies of natural proteins, so new tracks can be translated back into proteins to see what their properties will be.

Proteins have different properties, such as weight, size, method of laying and others. In addition to these trivial traits, they have another one – proteins fluctuate. Like any other molecules, when heated, they tremble finely with their characteristic frequency. The frequency depends on their composition – each amino acid also vibrates in its own way. And where there is a frequency, there may be sound.

Scientists from the Massachusetts Institute of Technology made music out of proteins and wrote an article about it. To write music, they went through several stages. 

First, they translated the frequency of amino acid vibrations into music. They analyzed the structure of all 20 amino acids forming proteins and obtained a set of frequencies with which they normally fluctuate. After that, they translated these frequencies into those that can be heard by multiplying the lowest frequency, tyrosine, by so many times to get si sub-octaves (61.7 Hz). They did the same with the rest of the amino acids – they do it in a similar way in ordinary music if they want to transpose it, that is, translate a piece from one key to another.

It turned out that tyrosine sounds the lowest, and alanine sounds the highest, its frequency is 20,000 Hz. It should be noted here that each amino acid emits many sounds at different frequencies at once, but the analysis showed that for each amino acid there is the most appropriate tone for it. Scientists have decomposed amino acids into notes according to which note the sound of each amino acid is closer to. 

Then they used data on the secondary structure of the protein to determine the volume and duration of the notes. They decided that disordered secondary structures would mean long notes, β-folds would be short, and α-spirals would be very short. In addition, the β-folds should be louder and everything else quieter. After that, the scientists voiced sections of several proteins – lysozyme, myoglobin, amyloid and others.

You can listen to the "music of proteins" in the MIT press release Translating proteins into music, and back - VM.

After that, the authors trained the neural network using three sets of protein fragments rich in β-folded structures, with a predominance of α-helices, and a third, where everything is equal. After training, the neural network was able to come up with protein melodies, which can also be translated back into proteins.

Scientists noticed that the neural network writes proteins with characteristic motifs that it "heard" in training sets. This means that the neural network has learned the principles of protein design and with its help it will be possible to collect proteins of the composition and shape that are needed. Scientists believe that such a neural network can be given several enzymes to listen to, so that it then composes some with similar properties, but completely new.

In addition, scientists believe that with the help of their idea, it will be possible to explain to a wide audience the structure of the protein, the origin of diseases associated with improper folding of molecules, and let people listen to how mutations sound. Finally, the authors write that their method makes it possible to better understand the relationship between sound and matter. 

Portal "Eternal youth" http://vechnayamolodost.ru