AI's Effects on Protein Folding

Artificial Intelligence (AI) is transforming our understanding of biology in ways that would have shocked scientists decades ago. One particular standout evolution of our understanding is its impact on protein folding. Proteins are one of the core four macromolecules that drive nearly every process in most living organisms. An example of their usage can be seen with the protein hemoglobin, which acts as an oxygen carrier in human red blood cells. The journey of proteins begins at the primary structure with a chain of amino acids, but eventually these chains of amino acids will fold into tertiary structures, enabling them to function properly. For decades, our understanding of how proteins fold was a driving question begging to be unraveled, often referred to as the “protein folding problem.” Regarding this difficult question, AI systems have been proposed to aid scientists’ approach to this challenge with promising implications in fields such as medicine and global health.

For those of you reading who study biology, the mantra that shape dictates function is likely engrained within your mind. Protein folding matters due to that mantra remaining true, structure determines function. As a result, if a protein folds incorrectly, it may lose its core functions or even become malignant. Common diseases linked to this phenomenon are Alzheimer’s, Parkinson’s, cystic fibrosis, and some forms of cancer. In previous stages of imaging, to determine a protein’s structure, many researchers and scientists relied on techniques deemed experimental, such as X-ray crystallography, nuclear magnetic resonance (NMR), spectroscopy, or cryo-electron microscopy. Quickly, many researchers began to understand that while these techniques offered a powerful way of obtaining desirable results, their limitations, such as costliness, time inefficiencies, and technical complexities, held back their overall appeal. This included analysis often going as far as some instances where solving the protein structure of a single protein would take months or years. These systematic limitations created a major knowledge gap in the field of biology, where the number of known protein structures quickly outpaced the number of experimentally determined structures.

Here is when Artificial Intelligence changes the story. AI has begun to close this seemingly impossible gap by learning and connecting patterns of amino acid sequences to tertiary structures. In the past, researchers relied on experimentally determining structure; however, AI models analyze massive datasets of the known protein structures previously determined to predict how new proteins will fold, potentially proposing cures to some diseases. In a current well-known example, AlphaFold, an AI system developed through DeepMind, demonstrated a breakthrough in accuracy benchmarks. In 2020, AlphaFold displayed this claim in the Critical Assessment of Structure Prediction (CASP), a global competition that determines the best methods of figuring out protein folding behavior methods. The predictions provided a way to address the gap in this field by providing results that were extremely accurate, close to results obtained from experimental methods, and thus marking a turning point in biology.

In this specific example, AlphaFold works not only by being trained on multiple datasets, but by combining deep learning techniques with biological knowledge. Since proteins interact with each other in the three-dimensional space, AlphaFold uses its neural networks, trained on thousands of known protein structures, to recognize these interactions. Contrary to popular belief, it does not simply use an algorithm or guess, but instead, it predicts distances and angles between amino acids to assemble those predicted interactions into a map. Moreover, in 2021, DeepMind released AlphaFold2, sending a shockwave through the research field when they also released the predicted structures for nearly every human proteome to be publicly available, along with other organisms. The reason for this shockwave was that the information was not kept behind a paywall or any other limitations, rather it was made open access and thus allowed scientists to explore protein structures without needing specialized equipment. Furthermore, beyond medicine, AI protein folding remains as a strong standing solution to fields even outside of biology, such as biotechnology and public health.

Connecting this rather astonishing discovery to HOSA members’ lives, we can see this intersection of AI and protein folding as a representation of an exciting development for the future of healthcare. For future physicians, researchers, and health leaders, many might struggle with this idea of how a complex task can be forged to an efficient path, but responsibility remains on us to determine how these AI-generated insights are produced and how we can evaluate their reliability. By solving a keystone problem in how common diseases are linked to proteins, AI has shown its potential to complement how humans discover and innovate rather than replace it. In other words, exposure to these technologies now rather than later can aid students in preparing for careers that will eventually become synonymous with Artificial Intelligence.

Thomas Bezza

Representative

2026