Molecular Mechanics Force Field Optimisation Using Parallel Computers

Edgington, Paul Rex (1992) Molecular Mechanics Force Field Optimisation Using Parallel Computers. PhD thesis, University of Glasgow.

Abstract

Ever since the discovery that the bond lengths and angles between the same atomic types vary little between different molecules chemists have been using models of molecules to aid them in their research. This technique is known as molecular modelling and is now usually performed with the aid of computers. One of the main areas of interest in molecular modelling is that of molecular mechanics calculations. Molecular mechanics (MM) is the use of empirical equations and their associated parameter to determine the energy of any particular molecular conformation and the use of that energy in helping to predict such things as the minimum energy conformation. The equations and the parameters are together described as the MM force field. One drawback of molecular mechanics is that in certain situations where its application would be useful there is an absence of reliable parameters so giving rise to the situation where calculations are often performed using 'guesstimates' for parameter values. To improve this situation an investigation into the possible forms of a general purpose compact force field was undertaken. There exist several force fields that reduce the number of parameters required by simplification of the situations where a parameter may be applied, for example a torsional barrier would be defined in terms of the central two atom types only rather than the more rigorous case where the atom types of all four atoms in the torsion angle would be used to decide which parameter is to be used. Some work had also already been done on reducing the number of parameters by calculating some by using other empirical formulae, an example of this being the calculation of bond stretch constants from their associated bond length parameters. These equations usually require some optimisable 'constants' but, in general, the technique resulted in an overall reduction in the number of parameters that are required to be optimised. What is not obvious is which of these methods are best at reducing the number of parameters without having an excessive effect on the accuracy of the results produced by the final force field. For each of these published force fields it is possible to adapt an energy minimisation program to calculate the minimum energy structures of known molecules and then compare the calculated results with experimental data. This will give us an overview of the accuracy for that force field but will not show which of the methods used within that force field have been the most effective. As MM is an empirical approach there are force fields which have basically the same form but have different parameters and it is usually unwise to interchange the parameters between them without reoptimising the whole force field. At the start of the project the methods used to optimise force fields were slow and the production of even an optimised force field for alkanes could take many years even after the form of the equations had been decided. It was obvious that using normal methods of optimisation it would take far too long to implement a system to vary the methods of parameter reduction and then reoptimise the force field in each case to see its effects. To overcome this problem it was decided to produce the required force fields by computer optimisation. This was made possible in mainly by the recently available parallel computing power of the Inmos transputer chip. The method chosen the controlling computer program to alter the parameters and so could be left for long periods to produce an optimised force field in a fraction of the time that would previously have been required. The first studies were done on small sets of alkene data as initially the processing power was limited. These initial studies culminated in investigations using a set of 50 alkene structures. These showed that a highly reduced force field is a viable option for alkenes, however alkenes are not very representative of all the atom types that will be needed so it was decided to introduce some more atom types before deciding on the final form of the force field. To this end a set of 109 structures was constructed which contained the following atom types: H, Csp2, Csp3, Osp2, Cl, Br and F. After considering the results of optimisations with this set of structures the form of a highly reduced parameter force field was decided upon. This force field was used in a limited study using a set of 243 structures with 25 different atom types. The following table summarises the results of this optimisation. Conformer refers to energy difference measures between conformers, 'Average Diff' is the average difference between the experimental value and that predicted by the force field and 'Average Error' is the average experimental error for that property type. Although not totally conclusive this study indicated that a highly reduced force field would be a valuable addition to the range of force fields available to the molecular modeller, as while being not as accurate a fuller force field it would have the significant advantage of covering, with reasonable accuracy, those situations that are not parameterised in the more specific force fields.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Adviser: David White
• Keywords: Computational chemistry, Physical chemistry, Molecular chemistry
• Date of Award: 1992
• Depositing User: Enlighten Team
• Unique ID: glathesis:1992-76372
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 19 Nov 2019 15:09
• Last Modified: 19 Nov 2019 15:09
• URI: https://theses.gla.ac.uk/id/eprint/76372