Polymorphisms in immune-rerated proteins and viral spike proteins are high and complicate host-virus interactions. Therefore, diversity analysis of the protein structures is essential to understand the mechanism of the immune system. However, experimental methods, including X-ray crystallography, nuclear magnetic resonance, and cryo-electron microscopy, have several problems: (i) they are conducted under different conditions from the actual cellular environment, (ii) they are laborious, time-consuming, and expensive, and (iii) they do not provide information on their thermodynamic behaviors. In this paper, we propose a computational method to solve these problems by using MD simulations, persistent homology, and a Bayesian statistical model. We apply our method to eight types of HLA-DR complexes to evaluate the structural diversity. The results show that our method can correctly discriminate the intrinsic structural variations caused by amino acid mutations from the random fluctuations caused by thermal vibrations. In the end, we discuss the applicability of our method in combination with existing deep learning-based methods for protein structure analysis.

免疫相关蛋白和病毒刺突蛋白的多态性很高，使宿主-病毒相互作用复杂化。因此，蛋白质结构的多样性分析对于理解免疫系统的机制至关重要。然而，包括 X 射线晶体学、核磁共振和冷冻电子显微镜在内的实验方法存在几个问题：(i) 它们是在与实际细胞环境不同的条件下进行的，(ii) 它们费力、耗时，并且昂贵，并且（iii）它们不提供有关其热力学行为的信息。在本文中，我们提出了一种计算方法，通过使用 MD 模拟、持久同源性和贝叶斯统计模型来解决这些问题。我们将我们的方法应用于八种类型的 HLA-DR 复合体以评估结构多样性。结果表明，我们的方法可以正确地区分由氨基酸突变引起的内在结构变异与由热振动引起的随机波动。最后，我们结合现有的基于深度学习的蛋白质结构分析方法，讨论了我们的方法的适用性。