By Zoe Stanley
There are several phrases deemed the “central dogma of biology” over time. An important one: “structure determines function”. The logic behind this, as is the logic behind all biology, is biochemistry. Proteins, the “actors” and “workers” of your cells, have very specialized functions, each determined by its exact function. When these structures change, the results can be catastrophic. Because of this, an important part of biology is understanding the structures of proteins, and how they come to be. Throughout scientific history, there have been many techniques developed to understand this. From nuclear magnetic resonance to X-ray crystallography, we have developed beautiful images of proteins, with varying levels of details. However, those methods of imaging often cost hundreds of thousands of dollars, and take long periods of time. Recently, an alternative has emerged: AlphaFold. Because protein structure is usually determined by the chemical interactions between its amino acids, it has long been postulated that, given sophisticated enough technology, scientists could accurately predict protein structure based only on the amino acid sequence[1]. Recently, this futuristic hope has come to fruition.
In 2021, a group of scientists published a paper in Nature asserting that they had developed a software capable of predicting protein structure with high accuracy to experimental observations. While even experimental methods rely on similar protein structures to provide an accurate model, AlphaFold has shown its capability to predict protein structures that have never been experimentally solved–indicating that AlphaFold’s programming uses just the biochemistry involved in the structure and not previously solved structures. It’s important to mention that AlphaFold is not the first such attempt. Until recently, most computational approaches have taken one of two paths; either only analyzing molecular dynamics, or only analyzing the evolutionary development of a protein over time[1]. AlphaFold combines both methods, and shows far better accuracy at both than its predecessors. Already, AlphaFold has been used to analyze proteins on the SARS-COV2 virus, allowing COVID-19 therapies to advance rapidly[2]–and this is only the start. AlphaFold and its implications raise many questions, many issues, and many future experiments.
Many a sci-fi movie has told us horror stories about technology to make the “perfect child”, or of the consequences of knowing one’s genetics before birth. But the fact is, we already do that. Amniotic screenings for birth defects and genetic diseases are commonly offered to pregnant people, and can often save or extend the lives of infants with deadly diseases by offering specialized healthcare even before birth. What if we could look further? If structure determines function, why can’t we take this revolutionary AI and ask it more questions? Questions, perhaps, about whether our proteins are shaped the right way–and what the consequences might be if they’re not.
Comments