Markov state models (MSMs) are a powerful approach for analyzing the long-time behaviors of protein motion using molecular dynamics simulation data. However, their quantitative performance with respect to the physical quantities is poor. We believe that this poor performance is caused by the failure to appropriately classify protein conformations into states when constructing MSMs. Herein, we show that the quantitative performance of an order parameter is improved when a manifold-learning technique is employed for the classification in the MSM. The MSM construction using the K-center method, which has been previously used for classification, has a poor quantitative performance.

中文翻译：

---

具有流形学习技术的马尔可夫状态模型的阶跃参数精确计算方法

马尔可夫状态模型（MSM）是使用分子动力学模拟数据分析蛋白质运动的长期行为的有力方法。但是，它们相对于物理量的定量性能差。我们认为，这种较差的性能是由于在构建MSM时未能适当地将蛋白质构象分类为状态而引起的。在这里，我们表明，在MSM中采用流形学习技术进行分类时，阶次参数的定量性能得到了改善。以前用于分类的使用K中心方法的MSM构造定量性能较差。