Researchers have made a major advancement in understanding protein stability by utilizing artificial intelligence (AI). The team employed AlphaFold2 to investigate how mutations influence protein stability—a key element in ensuring proper protein function and preventing diseases such as Alzheimer’s.
Researchers at the Center for Algorithmic and Robotized Synthesis within the Institute for Basic Science have made significant progress in comprehending protein stability through the use of AI technology. They utilized AlphaFold2 to study how mutations impact protein stability, which is essential for the proper functioning of proteins and for preventing diseases like Alzheimer’s.
DeepMind’s AlphaFold algorithm has revolutionized the field of biology by accurately predicting a protein’s structure based on its genetic information, thereby making structural biology more accessible. However, two critical questions still linger: Will the predicted structures fold correctly and maintain their shape? Additionally, there is a broader inquiry regarding how AI algorithms like AlphaFold actually operate.
A significant restriction of AlphaFold is that it was trained solely on a set of stable proteins that remain folded at typical physiological temperatures. Consequently, it forecasts the most probable folded structure without assurance that it will indeed fold correctly or remain stable. Understanding and foreseeing protein stability is vital because unstable proteins can misfold, leading to malfunctions and serious health issues, thus forcing cells to expend considerable energy to eliminate them. Furthermore, most proteins are generally marginally stable, making them very vulnerable to mutations that can cause them to unfold. This emphasizes that protein engineering involves navigating a complex terrain filled with dysfunctional protein sequences that have folding issues. Hence, the next phase of using AlphaFold should focus on predicting how mutations will alter stability.
An essential question explored in this study was whether AlphaFold has grasped the fundamental principles governing protein folding or if it merely functions as a high-dimensional regression model that identifies statistical trends. This inquiry pertains to the ability to generalize: if AlphaFold has truly learned the physical principles involved, it should perform effectively on new protein sequences it hasn’t encountered before.
The two IBS researchers, John MCBRIDE and Tsvi TLUSTY, specifically aimed to investigate this premise. They sought to determine whether AlphaFold could accurately predict the stability effects of mutations. Since there are vastly more possible mutations than training data points used for AlphaFold, even the most advanced regression models might not fully capture the entire spectrum of mutation effects. This task is quite challenging, as critical stability alterations often occur with minimal structural modifications that are difficult to anticipate. Nevertheless, there are promising signs within the structural changes predicted by AlphaFold that provide insightful clues about potential stability shifts.
The IBS researchers demonstrated this by comparing the structural alterations caused by mutations to the differences in stability measured between the original (wild-type) and mutated proteins [1]. A crucial element was employing a probe that could detect subtle changes effectively. They developed an innovative metric called effective strain [2], which identifies minor yet significant changes in protein structure linked to stability.
Upon examining thousands of mutations, they discovered that the effective strain metric aligns with the extent of stability changes. In other words, significant structural alterations (as predicted by AlphaFold) correspond with substantial changes in stability.
John MCBRIDE, the lead author, remarked, “This strongly indicates that the structures predicted by AlphaFold carry meaningful physical information, especially concerning stability. We must create new physical models to further interpret this information.”
These findings pave the way for enhanced protein engineering, which entails designing proteins for specific roles. By gaining deeper insights into how mutations influence stability, researchers can more adeptly explore the intricate landscape of protein design, potentially leading to breakthroughs in drug development and treatments for diseases linked to protein misfolding.
This research represents a significant milestone in the ongoing endeavor to use AI for deciphering biological complexities and highlights the necessity for additional studies to fully harness the capabilities of AI in scientific advancements.
