Planar Chirality for Acid/Base Responsive Macrocyclic Pillararenes Induced by Amino Acid Derivatives: Molecular Dynamics Simulations and Machine Learning.
Lulu Fu,Ranran Wang,Qiang Zhu,Yuming Gu,Lifeng Zheng,Yuan Chen,Juli Jiang,Jing Ma
DOI: https://doi.org/10.1021/acs.jctc.2c01265
2023-01-01
Journal of Chemical Theory and Computation
Abstract:Chirality is ubiquitous in nature, ranging from a DNA helix to a biological macromolecule, snail's shell, and even a galaxy. However, the precise control of chirality at the nanoscale is a challenge due to the structure complexity of supramolecular assemblies, the small energy differences between different enantiomers, and the difficulty in obtaining polymorphic crystals. The planar chirality of water-soluble pillar[5]arenes (called WP5-Na with Na ions in the side chain) host triggered by the addition of chiral L-amino acid hydrochloride (L-AA-OEt) guests and acid/base is rationalized by the relative stability of different chiral isomers, being estimated by molecular dynamics (MD) simulations and quantum chemical calculations. As an increase in the pH value, the change from a positive to a negative value of the free energy difference (ΔG) between two conformations, pR-WP5-Na⊃L-AA-OEt and pS-WP5-Na⊃L-AA-OEt, suggests an inversed preference of the pS-WP5-Na conformer induced by the deprotonated L-arginine ethyl ester (L-Arg-OEt) at pH = 14, which is supported by the circular dichroism (CD) experiments. On the basis of 2256 WP5-Na⊃L-Ala-OEt and 3299 WP5-Na⊃L-Arg-OEt conformers sampled from MD, the gradient boosting regression (GBR) model exhibits a satisfactory performance (R2 = 0.91) in predicting the chirality of WP5-Na complexations using host-guest binding descriptors, including the geometry matching and binding sites and modes (electrostatics and hydrogen bonding). The machine learning model also performs well on external tests of different hosts (using different side chains and cavity sizes) with the addition of 22 other different guests, with the average chirality prediction accuracy of ML versus experimental CD determinations of 92.8%. The easily accessible host-guest features, binding position coordination and size matching between the cavity and guest, exhibit a close correlation to the chirality of different macrocyclic molecules, water-soluble pillar[6]arenes (WP6) versus WP5, in complexation with different amino acid guests. The exploration of efficient host-guest features in ML displays the great potential of building a large space of various assembled systems and accelerating the on-demand design of chiral supramolecular systems at the nanoscale.