Abstract:
Proteins are organic compounds made up of chains of amino acids. Chemical and
physical properties determine the 3-dimensional structure and folding of a protein. A
protein needs to be folded into its proper 3D structure for its function to remain intact.
The protein structure prediction problem has real-world implication, since the 3D
structure of a protein gives important clues regarding its function, localization, and
interactions. Wet laboratory techniques are costly in terms of time and effort,
consequently having a right protein structure prediction model reduces cost and time
by eliminating some of the initial wet lab work. Consequently, we need to study
methods that predict protein structures. In this thesis, we present an improved scatter
search (SS) algorithm for predicting all-atoms protein structures using the
CHARMM22 energy model. Our algorithm produces a 3D structure of the whole
protein by minimizing the energy function linked to protein folding. This is based on
a sequence of amino acids as well as on data collected from known protein structures
for comparative purposes. Defined as an evolutionary algorithm, SS relies on a
population of candidate solutions. Candidate solutions, over a number of iterations,
experience evolutionary operations which combine intense search and diversification.
Our algorithm is evaluated on few proteins, whose structure is defined in a Protein
Data Bank (PDB). The results generated by the improved SS algorithm are compared
with those of other energy models. Our results showed that our algorithm produces
3D structures with good and promising root mean square deviations from the
reference proteins. This study also demonstrates the advantage of the CHARMM22
energy model.
