Department of Cell Biology, Microbiology and Molecular Biology, University of South Florida, Tampa, Florida 33620, USA.
Physics and Materials Science Research Unit, University of Luxembourg, 162a Avenue de La Fïancerie, Luxembourg City L-1511, Luxembourg.
J Chem Phys. 2020 Sep 7;153(9):094115. doi: 10.1063/5.0022058.
The reliability of molecular mechanics (MM) simulations in describing biomolecular ion-driven processes depends on their ability to accurately model interactions of ions simultaneously with water and other biochemical groups. In these models, ion descriptors are calibrated against reference data on ion-water interactions, and it is then assumed that these descriptors will also satisfactorily describe interactions of ions with other biochemical ligands. The comparison against the experiment and high-level quantum mechanical data show that this transferability assumption can break down severely. One approach to improve transferability is to assign cross terms or separate sets of non-bonded descriptors for every distinct pair of ion type and its coordinating ligand. Here, we propose an alternative solution that targets an error-source directly and corrects misrepresented physics. In standard model development, ligand descriptors are never calibrated or benchmarked in the high electric fields present near ions. We demonstrate for a representative MM model that when the polarization descriptors of its ligands are improved to respond to both low and high fields, ligand interactions with ions also improve, and transferability errors reduce substantially. In our case, the overall transferability error reduces from 3.3 kcal/mol to 1.8 kcal/mol. These improvements are observed without compromising on the accuracy of low-field interactions of ligands in gas and condensed phases. Reference data for calibration and performance evaluation are taken from the experiment and also obtained systematically from "gold-standard" CCSD(T) in the complete basis set limit, followed by benchmarked vdW-inclusive density functional theory.
分子力学 (MM) 模拟在描述生物分子离子驱动过程中的可靠性取决于其准确模拟离子与水和其他生化基团同时相互作用的能力。在这些模型中，离子描述符是根据离子-水相互作用的参考数据进行校准的，然后假设这些描述符也将令人满意地描述离子与其他生化配体的相互作用。与实验和高级量子力学数据的比较表明，这种可转移性假设可能会严重失效。提高可转移性的一种方法是为每个独特的离子类型及其配位配体对分配交叉项或单独的非键描述符集。在这里，我们提出了一种替代解决方案，直接针对错误源并纠正表示不当的物理。在标准模型开发中，配体描述符从未在离子附近存在的高电场中进行校准或基准测试。我们为一个代表性的 MM 模型证明，当其配体的极化描述符得到改进以响应低场和高场时，配体与离子的相互作用也会得到改善，并且可转移性误差会大大降低。在我们的案例中，整体可转移性误差从 3.3 kcal/mol 降低到 1.8 kcal/mol。这些改进是在不影响配体在气相和凝聚相中的低场相互作用准确性的情况下实现的。校准和性能评估的参考数据来自实验，也系统地从完全基组限制的“黄金标准” CCSD(T) 获得，然后进行基准测试的范德华包容性密度泛函理论。
