University of Chicago , 5801 S. Ellis Ave. , Chicago , Illinois 60637 , United States.
University of Notre Dame , Notre Dame , Indiana 46556 , United States.
J Chem Theory Comput. 2018 Jun 12;14(6):2881-2888. doi: 10.1021/acs.jctc.8b00192. Epub 2018 May 14.
We present a seamless coupling of a suite of codes designed to perform advanced sampling simulations, with a first-principles molecular dynamics (MD) engine. As an illustrative example, we discuss results for the free energy and potential surfaces of the alanine dipeptide obtained using both local and hybrid density functionals (DFT), and we compare them with those of a widely used classical force field, Amber99sb. In our calculations, the efficiency of first-principles MD using hybrid functionals is augmented by hierarchical sampling, where hybrid free energy calculations are initiated using estimates obtained with local functionals. We find that the free energy surfaces obtained from classical and first-principles calculations differ. Compared to DFT results, the classical force field overestimates the internal energy contribution of high free energy states, and it underestimates the entropic contribution along the entire free energy profile. Using the string method, we illustrate how these differences lead to different transition pathways connecting the metastable minima of the alanine dipeptide. In larger peptides, those differences would lead to qualitatively different results for the equilibrium structure and conformation of these molecules.
我们提出了一套代码的无缝耦合，这些代码旨在进行高级采样模拟，并与第一性原理分子动力学（MD）引擎相结合。作为一个说明性的例子，我们讨论了使用局部和混合密度泛函（DFT）获得的丙氨酸二肽的自由能和势能面的结果，并将其与广泛使用的经典力场 Amber99sb 的结果进行了比较。在我们的计算中，使用混合泛函的第一性原理 MD 的效率通过分层采样得到提高，其中使用局部泛函获得的估计值来启动混合自由能计算。我们发现，从经典和第一性原理计算得到的自由能面不同。与 DFT 结果相比，经典力场高估了高自由能态的内能贡献，并且低估了沿整个自由能曲线的熵贡献。使用字符串方法，我们说明了这些差异如何导致连接丙氨酸二肽亚稳最小值的不同跃迁途径。在更大的肽中，这些差异将导致这些分子的平衡结构和构象产生定性不同的结果。
