قالب وردپرس درنا توس
Home / Science / Generate adversarial networks, generate functional protein sequences

Generate adversarial networks, generate functional protein sequences



ProteinGAN: Generate adversarial networks and generate functional protein sequences

This figure summarizes the training of ProteinGAN. Given a random input vector, the generator network will generate a protein sequence, and the discriminator network will score it by comparing it with the natural protein sequence. The generator tries to trick the discriminator by generating a sequence that ultimately looks like a real sequence (the generator never actually sees the real enzyme sequence). Image source: Repeka et al.

Proteins are large, highly complex, and naturally occurring molecules can be found in all living organisms. These unique substances are composed of amino acids, which are linked together by peptide bonds to form long chains, which can have a variety of functions and properties.

The specific sequence of arranging different amino acids to form a given protein ultimately determines the 3D structure, physical and chemical properties and molecular functions of the protein. Although scientists have been studying proteins for decades, so far, designing proteins that trigger specific chemical reactions has proven to be very challenging.

Researchers at Vilnius University in Lithuania and Biomatter Designs at Chalmers University of Technology in Sweden recently developed ProteinGAN, a generative adversarial network (GAN) that can process and “learn”

; different natural protein sequences.This unique network was published in Natural machine intelligence, And then use the information it obtains to generate new functional protein sequences.

Aleksej Zelezniak, an associate professor at Chalmers University of Technology, who led the research, told Phys.org: “Proteins are a series of amino acid sequences that make processes happen in all living systems and induce humans.” Including countless products from washing powders to anti-cancer and coronavirus treatments. Proteins are made up of 20 kinds of amino acids, which are arranged in different orders, the order of which determines the function of the protein.”

Creating a functional protein sequence is a very challenging task, because even a slight change in a given sequence can render the protein nonfunctional. Non-functional proteins may have harmful and undesirable effects, such as causing cancer or other diseases in humans or animals.

Zelezniak said: “If you want to make proteins consistent with human needs, he/she needs to correctly understand the sequence of amino acids and the given astronomical numbers for making these proteins. This is not an easy task.” Inspired by the latest developments, especially the generation of realistic photos and videos, we would like to know whether the current AI technology is ready to produce the most complex molecule known to man-protein.”

The ProteinGAN model developed by Zelezniak and his colleagues is based on a well-known machine learning method called adversarial learning. Adversarial learning can be seen as a game “played” by two or more artificial neural networks. The first of these networks is called a “generator”, which produces specific types of data (for example, images, text, or protein sequences in the case of ProteinGAN). The second network is called the “discriminator” and it tries to distinguish between artificial data (such as protein sequences) created by the “generator” and real data or real data.

Subsequently, the generator uses the feedback provided by the discriminator (that is, the characteristics that allow it to distinguish between the generated data and the real data) to generate new data. The generator never processes or analyzes the actual data and the data it generates. Therefore, its learning depends only on the results of the analysis performed by the discriminator.

Zelezniak said: “By repeating this process repeatedly, both networks have become better at doing things until they can’t distinguish the generated sequence from the real sequence.” “Using the AI ​​tool we developed, we can generate active but not Functional proteins that exist or have not yet been discovered.”

In preliminary experiments conducted by the researchers, ProteinGAN produced a new highly diverse protein sequence whose physical properties are similar to those of natural protein sequences. Zelezniak and his colleagues used malate dehydrogenase (MDH) as a template enzyme, showing that many sequences produced by ProteinGAN are soluble and exhibit MDH catalytic activity, which means they may be interesting in medical and research environments Applications. In the future, ProteinGAN can be used to discover new protein sequences with different characteristics, which may have important value for a variety of technologies and scientific applications.

Zelezniak said: “Our research laboratory focuses on AI technology based on synthetic biological applications.” “We are currently working to solve emerging problems such as plastic pollution. I believe that artificial intelligence will help build better organisms. To adapt to this particular problem.”


Unique AI method can generate protein to accelerate drug development


More information:
Use generative adversarial networks to expand the sequence space of functional proteins. Natural machine intelligence(2021). DOI: 10.1038 / s42256-021-00310-5.

©2021 Science X Network

Citation: ProteinGAN: Generate adversarial networks and generate functional protein sequences (April 2, 2021). The network will be available from https://phys.org/news/2021-04-proteingan-adversarial- on April 4, 2021. network-functional-protein.htmlSearch

This document is protected by copyright. Except for any fair transactions for private learning or research purposes, no content may be copied without written permission. The content is for reference only.




Source link