Answers are grounded in this paper and cite their sources.
The 3D structure of proteins is directly linked to their function, making its determination crucial for understanding biological processes and addressing issues related to human health and life sciences. Despite the continuous experimental acquisition of new protein structures, there remains a significant gap between the number of protein sequences available and those that have an established experimental high-resolution tertiary structure. Several computational approaches have focused on predicting protein structures using either templates or empirical force field modeling. In recent years, various methods have been combined to address the individual limitations of these approaches, leading to the development of AlphaFold 3 (AF3) by Google DeepMind. AF3 enables prediction of 3D protein structures with high accuracy based on its amino acid sequence. In this study, we benchmarked applicability, performance, and limitations of AF3 for predicting 3D structure of a broad series of proteins, including SARS-CoV-2 coronavirus proteins, other bacterial and viral proteins, as well as some plant enzymes. We found that AlphaFold 3 could capture the overall backbone features of the most examined proteins in terms of small deviation from available X-ray structures. Some minor miss-folding of N- and C-terminal segments were common, which, often, did not affect biological roles of the studied proteins. In cases involving protein dimers or higher-order oligomers, there are notable differences between the predicted AF3 models of a single-chain monomer and their corresponding experimental structures. These discrepancies are particularly evident in regions related to protein dimerization, assembly, and binding interfaces. Ultimately, while capturing the overall fold, predicting the complex structure of the Spike glycoprotein is still beyond the current capabilities of AF3.
No takes yet. Share an insight, caveat, or question.
Duma et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: