Generative AI Helps Design New Proteins
The amino acid chains that make up proteins are folded into three-dimensional forms, which determine the function of the protein. These shapes have evolved over billions and millions of years. They are complex and varied, but limited in numbers. Researchers have been able to create new folding patterns by studying how proteins fold.
Kim says that a big challenge has been to create folds which are both functional and possible. It’s hard to predict what folds are real and functional in a protein.
The researchers developed a new system called ProteinSGM that draws on a large number of images-like representations for existing proteins which encode their structure with accuracy. Researchers feed these images to a generative model that gradually adds noise, until each image is all noise. The model monitors how images get noisier, and then reverses the process to learn how to turn random pixels into images that are clear.
Source:
https://www.technologynetworks.com/tn/news/generative-ai-helps-design-new-proteins-372975