期刊
出版社
WILEY
DOI: 10.1002/wcms.1690
关键词
conformational search; cyclic molecules; inverse kinematics; machine learning; structure representation
Inverse kinematics provides a mathematically strict solution for handling cyclic molecules in conformational searches, allowing for the narrowing of conformational space and improving the accuracy and efficiency of models.
Conformational searches and ML-driven geometry predictions (e.g., AlphaFold) work in the space of molecule's degrees of freedom. When dealing with cycles, cyclicity constraints impose complex interdependence between them, so that arbitrary changes of cyclic dihedral angles lead to heavy distortions of some bond lengths and valence angles of the ring. This renders navigation through conformational space of cyclic molecules to be very challenging. Inverse kinematics is a theory that provides a mathematically strict solution to this problem. It allows one to identify degrees of freedom for any polycyclic molecule, that is, its dihedral angles that can be set independently from each other. Then for any values of degrees of freedom, inverse kinematics can reconstruct the remaining dihedrals so that all rings are closed with given bond lengths and valence angles. This approach offers a handy and efficient way for constructing and navigating conformational space of any molecule using either naive Monte-Carlo sampling or sophisticated machine learning models. Inverse kinematics can considerably narrow the conformational space of a polycyclic molecule to include only cyclicity-preserving regions. Thus, it can be viewed as a physical constraint on the model, making the latter obey the laws of kinematics, which govern the rings conformations. We believe that inverse kinematics will be universally used in the ever-growing field of geometry prediction of complex interlinked molecules.This article is categorized under: Structure and Mechanism > Molecular Structures Data Science > Artificial Intelligence/Machine Learning Data Science > Chemoinformatics
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据