The development of a new technology for designing proteins from scratch has completely changed how rapidly researchers can progress from conceptualising a protein to having it.
When researchers design a new protein, such as one used as a drug, they need to use a blueprint for assembling it.
Creating blueprints usually takes many months, with a researcher painstakingly figuring out each part of the drawing to ensure the desired structure.
A computer-generated model has now superseded this time-consuming work. What used to take months now takes one workday, and this could completely revolutionise how researchers make proteins.
Within the past 3 years, the researchers behind this potential revolution have published 11 articles in Science, four in Cell and 12 in Nature. The last of these articles was recently published in Nature.
“This is really exciting, because this method makes creating blueprints for designing proteins easier, and we can use the method for many purposes,” explains Christoffer Norn, Postdoctoral Fellow, Department of Biology, University of Copenhagen and Baker Lab, Institute for Protein Design, University of Washington, Seattle, USA.
Computer program assembles proteins
The computer program underpinning this revolution is trRosetta, a deep neural network trained to predict how amino acids act when they are assembled into a protein.
trRosetta is trained to predict the most likely structure of an amino acid sequence and which protein will emerge if the amino acids are assembled in a specific way.
Researchers enter a desired protein structure into the computer program, and it tests the possibilities one amino acid at a time. If a specific amino acid increases the likelihood of a desired result, trRosetta retains the amino acid and moves on to the next one. After perhaps 30,000 attempts, the program has an idea of the final blueprint for building a protein with exactly the properties the researchers have requested.
“Traditionally, you describe what the protein should look like, and then you try to make a blueprint by assembling many known protein structures. With this new method, we do not have to define what the protein should look like but instead ask the neural network to create numerous structures that can be stable and fulfil the desired function,” says Christoffer Norn.
Neural network accelerates protein design
Christoffer Norn and his colleagues have used the neural network to design six new protein functions in the past 6 months.
“This is really much faster than the traditional methods. It used to take months to make a new blueprint to design a protein from scratch. Today I can make one in 8 hours,” he says.
Christoffer Norn and his colleagues anticipate using the method to design new enzymes and proteins that can help treat various medical conditions.
The method is part of a larger revolution in designing proteins and predicting structure.
“The new methods for predicting structure accelerate not only the creative process of making new blueprints but also their validation. Testing new proteins typically used to take 6 to 12 months. Now we can get a good indication of whether a protein will behave the way we want in a few minutes,” explains Christoffer Norn.
Billions invested in protein design start-ups
trRosetta and similar programs for designing protein structures can take protein research and drug development to a new level.
Researchers can use this technology to design completely new proteins or ones that overcome some of the problems associated with the proteins used as drugs today.
For example, researchers would like to improve the drugs intended for some targets in the gastrointestinal tract.
Today, getting protein-based drugs from the mouth to the gastrointestinal tract is very difficult because the digestive proteases break down the proteins.
Researchers can potentially use trRosetta to design proteins that resist proteases and can thereby travel intact through the digestive system and to the intestines, where they can exert their effect.
“We can adjust the properties of the proteins and make them bind to target structures. We can thereby design drug candidates that bind to various receptors or molecules in the body and either activate them or deactivate them. There are many opportunities that now make progressing from an idea to the finished protein easier,” says Christoffer Norn.
Christoffer Norn is convinced that many new proteins will be developed soon using this method.
Nine start-up companies have already emerged from the Institute for Protein Design at the University of Washington, where Director David Baker is pioneering the technology. The companies have attracted investment totalling USD 3 billion.