In a significant leap for medicine and biotechnology, The University of Texas at Austin has unveiled an artificial intelligence model capable of crafting proteins for new drugs and vaccines. This model, known as EvoRank, taps into the mechanisms behind evolution to assist in the development of updated protein-based applications. Researchers highlighted the AI's capabilities at the International Conference on Machine Learning and detailed their findings in an article for Nature Communications.
Confronting the challenge of insufficient data that hampers AI's ability to effectively learn about proteins, EvoRank utilizes a vast array of naturally occurring proteins to bridge the gap. Daniel Diaz, a leading researcher and part of the Deep Proteins team at UT, described the process by which EvoRank operates. “Nature has been evolving proteins for 3 billion years, mutating or swapping out amino acids and keeping those that benefit living things", Diaz told UT News. He explained that EvoRank learns how to rank the evolution that we observe around us which in turn, informs the creation of new proteins.
The drive to synthesize and engineer proteins is not without its complications. Traditional approaches involve laborious trial and error, hindering the pace at which protein-based drugs are developed. Oftentimes, the daunting process requires over a billion dollars and can span more than a decade. The introduction of EvoRank has the potential to streamline this effort, rendering the design of biotechnologies not only faster but also more cost-effective. It's a development that could have a huge impact on an industry poised for substantial growth over the next ten years.
UT Austin, recognized for its leading AI research programs, has recently been allotted close to $2.5 million in grants to apply AI to protein engineering for vaccines targeting herpesviruses. Adam Klivans, who leads both IFML and Deep Proteins, emphasized the significance of AI in this arena. "Engineering proteins with capabilities that natural proteins do not have is a recurring grand challenge in the life sciences," Klivans stated, pointing out how perfectly suited generative AI models are for such tasks. Unlike AlphaFold by Google DeepMind, which predicts protein structures, the Deep Proteins group's EvoRank is designed to suggest alterations for specific functionalities.
The practical implications of this research are evident, with vaccine-maker Jason McLellan's lab at UT synthesizing viral proteins based on designs provided by the AI. Reflecting on the results, McLellan stated, "The models have come up with substitutions we never would have thought of," revealing the AI's knack for identifying previously unknown stabilizing modifications. This tool not only advances our understanding of protein stability and functionality but also holds the promise of reshaping the costly landscape of drug development.
