Department of Chemistry, University of Pavia, Pavia 27100, Italy.
ACS Chem Biol. 2024 Oct 18;19(10):2141-2143. doi: 10.1021/acschembio.4c00376. Epub 2024 Sep 24.
Recent advancements in AI-driven computational modeling, especially AlphaFold2, have revolutionized the prediction of biological macromolecule structures. AlphaFold2 enabled accurate predictions of structural domains and complex arrangements. However, computational models lack a clear metric for accuracy. This study explores whether computational models can match the crystallographic resolution of crystal structures. By comparing distances between atoms in models and crystal structures using tests, it was found that AlphaFold2 models are comparable to high-resolution crystal structures (1.1 to 1.5 Å). While these models exhibit exceptional quality, their accuracy is lower than the crystal structure with resolutions better than 1 Å.
近年来，人工智能驱动的计算建模的进步，特别是 AlphaFold2，彻底改变了生物大分子结构的预测。AlphaFold2 能够准确预测结构域和复杂排列。然而，计算模型缺乏明确的准确性衡量标准。本研究探讨了计算模型是否可以与晶体结构的晶体学分辨率相匹配。通过使用 测试比较模型和晶体结构中原子之间的距离，发现 AlphaFold2 模型可与高分辨率晶体结构（1.1 至 1.5Å）相媲美。虽然这些模型表现出了卓越的质量，但它们的准确性低于分辨率优于 1Å 的晶体结构。
