Scientists are using artificial intelligence (AI) to explore new uses in the development of cancer treatments, vaccines for new strains of influenza, and drugs that help patients tolerate gluten.
This type of protein is generated using a model similar to AI DALL-E, which creates images from text and uncovers capabilities of the human body that were previously unknown.
This type of protein is generated using a model similar to AI DALL-E
A year ago, the AI research company OpenAI announced the DALL-E model, capable of creating artwork based on keywords. Across social media platforms, keywords like “creative AI,” “AI art” consistently rank among the top searches, with AI-generated artworks emerging “like mushrooms after rain.” This technology has sparked considerable controversy, as it poses a threat to traditional artists and raises concerns about intellectual property rights.
By leveraging techniques similar to DALL-E and other AI image generation tools, scientists have created design models for a new type of protein that can alter the body’s normal metabolic processes.
Unlimited AI-Generated Protein Designs
According to the New York Times, the human body has around 20,000 genes coding for proteins, responsible for everything from digesting food to distributing oxygen throughout the body. Among these, the new type of protein recently discovered by scientists is not naturally occurring but has the potential to combat diseases and perform functions that the body cannot achieve on its own.
David Baker, director of the Institute for Protein Design at the University of Washington, has been researching this compound for over 30 years and successfully proved its viability in 2017. However, he did not anticipate that recent AI technology would accelerate research progress, reducing the time required for design prototypes from several years to just a few weeks.
David Baker and his team have created a new protein model using AI. (Image: New York Times).
“What we need are new types of proteins that can address diseases like cancer or pandemics. Thanks to AI, we can quickly create these proteins with higher success rates and more complex molecules for treatment,” Baker stated.
According to researcher Nate Bennett from the University of Washington, one of DALL-E’s strengths is its ability to execute any request made by humans. “From a simple command, it can generate an infinite number of different design models,” he explained.
To create images, DALL-E uses a neural network system based on the human brain’s neural networks. This network has been applied to help smartphones recognize user commands, assist self-driving cars in avoiding pedestrians, and translate foreign languages on Skype.
The neural network learns through an enormous digital dataset. For example, to recognize a dog, this AI must train on thousands of images of dogs. With DALL-E, scientists built the neural network by having the AI analyze millions of images along with text descriptions for each image.
As a result, AI can recognize the relationship between images and language. When users describe their requests to DALL-E, the neural network gathers the necessary individual properties and combines pixels from thousands of different images to create a complete image.
Researchers at the University of Washington applied a similar approach to generate a new type of protein. By leveraging the ability to predict the 3D shapes of any protein in the human body, Baker’s team created a complete design model for a protein that has never been seen in nature. Their goal is to produce proteins with various shapes, each serving a specific function such as destroying coronaviruses or treating cancer.
Challenges in Applying AI to Research
While DALL-E can link images and text to produce results, this system can also connect scientists’ specific requests regarding the type of protein they need and form design models for it. Researchers simply need to provide a request, and the AI tool will generate the corresponding 3D image.
“With DALL-E, if users can create an image of a panda eating bamboo, scientists can also ask the AI to create proteins that link together or form more complex connections,” said Namrata Anand, a former researcher at Stanford University.
Protein segments generated by AI. (Image: Namrata Anand).
However, the key difference is that humans can evaluate DALL-E’s creations with the naked eye, while protein structures cannot be judged in the same way. After AI generates a design, scientists must bring it to the laboratory to study its effects.
Therefore, some experts suggest that the application of AI technology in science should be considered carefully. “Creating new structures is just the beginning. The question is, what can this protein structure do?” said Professor Frances Arnold at the California Institute of Technology. Meanwhile, many other researchers argue that this new technology not only accelerates the pace of creating new proteins but also allows scientists to access previously unexplored inventions.
“The exciting thing is that they can not only experiment and discover the unexpected but also create freely while still meeting the intended purpose. AI helps them avoid the laborious trial and error with every type of protein on Earth,” said Jue Wang, a researcher at the University of Washington.
