MSGID: 1:317/3 645486d0   
   PID: hpt/lnx 1.9.0-cur 2019-01-08   
   TID: hpt/lnx 1.9.0-cur 2019-01-08   
    Researchers use generative AI to design novel proteins    
      
     Date:   
         May 4, 2023   
     Source:   
         University of Toronto   
     Summary:   
         Researchers have developed an artificial intelligence system that   
         can create proteins not found in nature using generative diffusion,   
         the same technology behind popular image-creation platforms such   
         as DALL-E and Midjourney.   
      
      
         Facebook Twitter Pinterest LinkedIN Email   
      
   ==========================================================================   
   FULL STORY   
   ==========================================================================   
   Researchers at the University of Toronto have developed an artificial   
   intelligence system that can create proteins not found in nature using   
   generative diffusion, the same technology behind popular image-creation   
   platforms such as DALL-E and Midjourney.   
      
   The system will help advance the field of generative biology, which   
   promises to speed drug development by making the design and testing of   
   entirely new therapeutic proteins more efficient and flexible.   
      
   "Our model learns from image representations to generate fully new   
   proteins, at a very high rate," says Philip M. Kim, a professor in   
   the Donnelly Centre for Cellular and Biomolecular Research at U of T's   
   Temerty Faculty of Medicine.   
      
   "All our proteins appear to be biophysically real, meaning they fold   
   into configurations that enable them to carry out specific functions   
   within cells."  Today, the journal Nature Computational Science published   
   the findings, the first of their kind in a peer-reviewed journal. Kim's   
   lab also published a pre- print on the model last summer through the   
   open-access server bioRxiv, ahead of two similar pre-prints from last   
   December, RF Diffusion by the University of Washington and Chroma by   
   Generate Biomedicines.   
      
   Proteins are made from chains of amino acids that fold into   
   three-dimensional shapes, which in turn dictate protein function. Those   
   shapes evolved over billions of years and are varied and complex, but   
   also limited in number. With a better understanding of how existing   
   proteins fold, researchers have begun to design folding patterns not   
   produced in nature.   
      
   But a major challenge, says Kim, has been to imagine folds that are both   
   possible and functional. "It's been very hard to predict which folds   
   will be real and work in a protein structure," says Kim, who is also a   
   professor in the departments of molecular genetics and computer science   
   at U of T. "By combining biophysics-based representations of protein   
   structure with diffusion methods from the image generation space, we can   
   begin to address this problem."  The new system, which the researchers   
   call ProteinSGM, draws from a large set of image-like representations of   
   existing proteins that encode their structure accurately. The researchers   
   feed these images into a generative diffusion model, which gradually   
   adds noise until each image becomes all noise. The model tracks how the   
   images become noisier and then runs the process in reverse, learning how   
   to transform random pixels into clear images that correspond to fully   
   novel proteins.   
      
   Jin Sub (Michael) Lee, a doctoral student in the Kim lab and first   
   author on the paper, says that optimizing the early stage of this   
   image generation process was one of the biggest challenges in creating   
   ProteinSGM. "A key idea was the proper image-like representation of   
   protein structure, such that the diffusion model can learn how to generate   
   novel proteins accurately," says Lee, who is from Vancouver but did his   
   undergraduate degree in South Korea and master's in Switzerland before   
   choosing U of T for his doctorate.   
      
   Also difficult was validation of the proteins produced by ProteinSGM. The   
   system generates many structures, often unlike anything found in   
   nature. Almost all of them look real according to standard metrics,   
   says Lee, but the researchers needed further proof.   
      
   To test their new proteins, Lee and his colleagues first turned to   
   OmegaFold, an improved version of DeepMind's software AlphaFold 2. Both   
   platforms use AI to predict the structure of proteins based on amino   
   acid sequences.   
      
   With OmegaFold, the team confirmed that almost all their novel sequences   
   fold into the desired and also novel protein structures. They then chose   
   a smaller number to create physically in test tubes, to confirm the   
   structures were proteins and not just stray strings of chemical compounds.   
      
   "With matches in OmegaFold and experimental testing in the lab, we could   
   be confident these were properly folded proteins. It was amazing to see   
   validation of these fully new protein folds that don't exist anywhere   
   in nature," Lee says.   
      
   Next steps based on this work include further development of ProteinSGM   
   for antibodies and other proteins with the most therapeutic potential,   
   Kim says.   
      
   "This will be a very exciting area for research and entrepreneurship,"   
   he adds.   
      
   Lee says he would like to see generative biology move toward joint   
   design of protein sequences and structures, including protein side-chain   
   conformations.   
      
   Most research to date has focussed on generation of backbones, the   
   primary chemical structures that hold proteins together.   
      
   "Side-chain configurations ultimately determine protein function, and   
   although designing them means an exponential increase in complexity, it   
   may be possible with proper engineering," Lee says. "We hope to find out."   
       * RELATED_TOPICS   
             o Matter_&_Energy   
                   # Biochemistry # Organic_Chemistry # Nature_of_Water #   
                   Civil_Engineering   
             o Computers_&_Math   
                   # Computer_Modeling # Mathematical_Modeling #   
                   Artificial_Intelligence # Computational_Biology   
       * RELATED_TERMS   
             o Computer_vision o Emerging_technologies   
             o Information_and_communication_technologies o   
             Electron_microscope o Technology o Artificial_intelligence o   
             Computing_power_everywhere o Radiography   
      
   ==========================================================================   
   Story Source: Materials provided by University_of_Toronto. Original   
   written by Jim Oldfield.   
      
   Note: Content may be edited for style and length.   
      
      
   ==========================================================================   
   Journal Reference:   
      1. Jin Sub Lee, Jisun Kim, Philip M. Kim. Score-based generative   
      modeling   
         for de novo protein design. Nature Computational Science, 2023;   
         DOI: 10.1038/s43588-023-00440-3   
   ==========================================================================   
      
   Link to news story:   
   https://www.sciencedaily.com/releases/2023/05/230504121014.htm   
      
   --- up 1 year, 9 weeks, 3 days, 10 hours, 52 minutes   
    * Origin: -=> Castle Rock BBS <=- Now Husky HPT Powered! (1:317/3)   
   SEEN-BY: 15/0 106/201 114/705 123/120 153/7715 218/700 226/30 227/114   
   SEEN-BY: 229/110 112 113 307 317 400 426 428 470 664 700 292/854 298/25   
   SEEN-BY: 305/3 317/3 320/219 396/45   
   PATH: 317/3 229/426