Automated Coarse-Graining of Biomolecules
Co-supervised by Jonathan Essex
A recent trend in molecular dynamics has been the development of coarse-grained (CG) models whereby multiple atoms are treated as a single particle. Simulations using CG models are typically 2-3 orders of magnitude more efficient than all-atom simulations at a cost of slightly reduced accuracy. Using these models it is then possible to investigate systems both much larger than and for much longer than previously, with multi-microsecond simulations representing hundreds of thousands of atoms now commonplace.
However, in order for these simulations to be performed, sets of parameters must be created, typically by comparison to all-atom simulations. Creating these parameter sets is often extremely time consuming and relies upon a degree of ‘chemical intuition’. This project aims to create a set of tools to assist in the creation of CG models by automating as much of the model development as possible. Using an all-atom molecular dynamics trajectory it is possible to perform a mapping to CG particles and calculate equilibrium bond properties and force constants. This is all that is required to parametrise the simplest class of CG models, but more complex models allow extensive adjustment of non-bonded parameters such as the coefficients of the Lennard-Jones potential or more rigorous electrostatics.
Along with these, tools are also being created for the analysis of membrane simulations for use in the Khalid group.
The ultimate aim of the project is to use these tools to produce a CG model of the class of membrane lipids Lipopolysaccharides (LPS). These are complex lipids found on the outer surface of Gram-negative bacteria with a varying number of hydrophobic tails (Lipid A), a core region of complex sugars and a long polysaccharide tail (O Antigen). The smallest LPS consist of just Lipid A and are around 300 atoms, whereas an LPS with a full O Antigen may contain over a thousand atoms. Due to their large number of atoms, complex structure and large number of polar hydroxyl groups it is believed that they will be a good target for a CG model with an accurate description of electrostatics. The ELBA (ELectrostatics BAsed) model has been developed in the Essex group for a range of simpler membrane lipids and allows CG beads to have both a partial charge and a point dipole and while it is less efficient than some other CG force fields it recovers some of the lost accuracy. ELBA is also well suited to multi-scale simulation where part of the system is simulated in all-atom detail while the surrounding region is coarse-grained. This allows high accuracy to be maintained in the region of interest while benefiting from the increased performance of CG models in the background.
The tools are available in a ‘work-in-progress’ state at https://bitbucket.org/jag1g13/cgtool
Systems and Synthetic Biology Modelling Group
School of Chemistry
University of Southampton
SO17 1 BJ