Computational prediction of the molecular conformation of druglike molecules and carbohydrates

  • Linghan Kong

Student thesis: Phd

Abstract

Molecular conformation is a crucial factor in determining the properties of a biomolecular system. Experimentally characterising molecular conformations can be challenging and time-consuming. This research aims to assess the feasibility of deploying the latest semiempirical quantum mechanical (SQM) and machine learning (ML) methods to model the energetics and conformations of druglike molecules and small carbohydrate molecules, and their potential to be applied to larger molecular systems. First, we built a dataset of 20 druglike molecules, of which 19 are FDA-approved drugs. The reference energies in this dataset were characterised using the coupled cluster method on density functional theory (DFT) geometries. We assessed the ability to reproduce reference conformational energies and geometries of GFN2-xTB, the best-performing SQM method, and ANI-2x, the latest implementation of the ML-based ANI methods, with our dataset. GFN2-xTB could provide overall reasonable conformational energies and geometries. However, it failed to predict correct conformational energy for conformers with π-π interactions. ANI-2x predictions had larger energetic errors than for GFN2-xTB, particularly for flexible molecules where long-range interactions may be neglected. Secondly, we deployed GFN2-xTB, ANI-2x and another ANI implementation fitted to coupled cluster energies, ANI-1ccx, to evaluate a conformational dataset for six monosaccharides using CCSD(T)-derived relative energies. We observed that ANI-1ccx reproduces the relative energies of all molecules with very good accuracy. ANI-2x yields slightly larger energetic errors than ANI-1ccx, whereas GFN2-xTB greatly overstabilised conformers relative to the 4C1 chair conformer. We found that ANI-2x significantly favours short hydrogen bonds, while GFN2-xTB underestimates both syn-diaxial repulsion and ¬endo-anomeric effects. Finally, we used a molecular mechanics (MM) force field to model the adhesion between a pristine graphene sheet and cellulose microfibril, an interlinked polysaccharide macromolecule, in the condensed phase. We observed the spontaneous rearrangement of the hydrogen bonding network within the cellulose microfibril upon its adhesion to graphene. We also observed progressive untwisting in the microfibril, due to interactions between the microfibril and the graphene sheet. The limitations of the approach is discussed, and we consider the potential for modelling large polysaccharide systems with hybrid SQM/MM and ML/MM methods.
Date of Award1 Aug 2023
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorNeil Burton (Supervisor) & Richard Bryce (Supervisor)

Keywords

  • graphene
  • molecular dynamics
  • machine learning
  • quantum mechanics
  • drug discovery
  • carbohydrates

Cite this

'