Insights into the Kinetic Partitioning Folding Dynamics of the Human Telomeric G-Quadruplex from Molecular Simulations and Machine Learning
Insights into the Kinetic Partitioning Folding Dynamics of the Human Telomeric G-Quadruplex from Molecular Simulations and Machine Learning
复制标题
从分子模拟和机器学习深入了解人类端粒 G-四链体的动力学分区折叠动力学
DOI：
10.1021/acs.jctc.0c00340
复制
发表时间：
2020-09-08
影响因子：
5.5
通讯作者：
Wang，Wei
中科院分区：
文献类型：
--
作者：
Bian，Yunqiang;Song，Feng;Wang，Wei
人类端粒DNA G-四链遵循动力学分配折叠机制。下面的折叠景观可能有许多被高自由能势垒隔开的极小值。然而，使用目前的理论模型来描述这种复杂的折叠景观仍然是一个具有挑战性的问题。在这项研究中，通过建立一个基于混合原子学结构的模型，该模型融合了杂交-1、杂交-2和椅型G-四链拓扑的结构信息，我们研究了涉及三个天然折叠的人端粒DNA的动力学分割折叠过程。该模型得到了验证，因为它再现了实验观察到的混合-1构象是主要折叠，并且混合-2构象在动力学上更容易接近。杂化-1构象的形成是三步机制，而杂化-2和椅型构象的形成是两步机制。同样，发现一类结构采用不适当的顺/反鸟嘌呤核苷酸组合的状态会极大地减缓折叠过程。此外，通过使用XGBoost机器学习算法，确定了3个原子间距离和6个二面角作为代表低维折叠景观的基本内部坐标。这种多盆地模型和机器学习算法相结合的策略可能对研究其他多态生物分子的构象动力学是有用的。
The human telomeric DNA G-quadruplex follows a kinetic partitioning folding mechanism. The underlying folding landscape potentially has many minima separated by high free-energy barriers. However, using current theoretical models to characterize this complex folding landscape has remained a challenging problem. In this study, by developing a hybrid atomistic structure-based model that merges structural information on the hybrid-1, hybrid-2, and chair-type G-quadruplex topologies, we investigated a kinetic partitioning folding process of human telomeric DNA involving three native folds. The model was validated as it reproduced the experimental observation that the hybrid-1 conformation is the major fold and the hybrid-2 conformation is kinetically more accessible. A three-step mechanism was revealed for the formation of the hybrid-1 conformation, while a two-step mechanism was demonstrated for the formation of hybrid-2 and chair-type conformations. Likewise, a class of state in which structures adopted inappropriate combinations of syn/anti guanine nucleotides was found to greatly slow down the folding process. In addition, by employing the XGBoost machine learning algorithm, three interatom distances and six dihedral angles were identified as essential internal coordinates to represent the low-dimensional folding landscape. The strategy of coupling the multibasin model and the machine learning algorithm may be useful to investigate the conformational dynamics of other multistate biomolecules.