Proteins are microscopic molecules that dictate everything from the behavior of viruses and bacteria to the regular operation of the human body. These tiny catalysts are made of a unique string of amino acids that determine both their 3D structure and function. Researchers have worked for decades to predict how proteins fold into these unique shapes to then build muscle or regulate hormones, among many other key processes. 

These complex shapes can also inform the ways antibodies spot and react to diseases attacking our immune systems. With that in mind, researchers have been working to find or engineer proteins that can then be used as powerful tools in treatments such as drug development or to improve our basic understanding of cell function.Christopher Snow, a professor in the Department of Chemical and Biological Engineering, said discovering and then outlining a protein’s shape was a time intensive process in the past.

However, AI tools now allow researchers to quickly identify and predict how proteins may interact and specifically link together like a lock and key with other molecules for a given purpose or need. New AI tools can also help guide researchers to genetically design proteins that may not even exist in nature but would be useful, said Snow. I have always tried to embrace the latest technology in my lab, and AI is clearly changing the field of biomolecular engineering, he said. The combination of tools available to us presents an exciting opportunity to make the research and identification processes more effective.

Snow said Google DeepMind’s AlphaFold2 first broke barriers in 2020 and is still the most popular and effective AI tool for protein structure prediction today. Depending on the size of the target, high-confidence models can be made in minutes with the system about a given protein’s structure when it folds. Since then, third generation methods, such as AlphaFold3 and its competitors, have emerged and extend the predictions to non-protein components like small molecules and nucleic acids. In many cases, the predictions are good enough to enable understanding without time-consuming real-life experimentation.

Source: https://source.colostate.edu/ai-protein-engineering/