Creating Knowledgeable Atoms for the Molecular Dynamics Simulations of Peptides

  • Matthew Burn

Student thesis: Phd


Computers are as fast as they have ever been, allowing the computation of phenomena that may have been considered a dream only a couple of decades ago. For biomolecular simulations, the increase in computing power is allowing larger and longer simulations than were ever possible before. Although computing power has come a long way, large scale ab initio simulations are still infeasible therefore requiring the use of force fields which have been shown to be unreliable in accurately representing realistic molecular dynamics. Force field design has always struggled with the trade-off between speed and accuracy, consequently a new approach should be considered. FFLUX is a novel force field based upon Gaussian process regression (GPR) machine learning models to learn atomic properties from ab initio data and used within polarisable and flexible atomistic simulations. This thesis details the development of methods to improve the production of atomistic GPR models through the use of active learning. FFLUX relies solely on GPR models to provide accurate predictions in order to accurately simulate the dynamics of a system, therefore the quality of the GPR models is of the utmost importance. Active learning is a method of iteratively improving a machine learning model by adding new data that will decrease the prediction errors of the model. The implementation of active learning is carried out by the in-house program ICHOR and the GPR by FEREBUS, both of which are discussed in-depth throughout this thesis. The application of GPR models are demonstrated alongside extensive analyses providing a solid foundation for future developments in FFLUX.
Date of Award31 Dec 2022
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorPaul Popelier (Supervisor) & Magnus Rattray (Supervisor)


  • Quantum Chemical Topology
  • Active Learning
  • Quantum Chemistry
  • Gaussian Process Regression
  • Force Field
  • Machine Learning
  • Computational Chemistry

Cite this