Modeling Proteins For Their Ground State Conformations And Their Secondary Structure Predictions

This thesis is concemed with the modeling of proteins and consists of two parts; computing for their ground state conformations and computing for their secondary structure predictions.To handle the first part of the thesis,we propose a genetic algorithm with a dynastic change strategy to determine the ground state conformations of proteins.The toy model which is a 2-D,2-rigid monomer model is adopted to handle the protein folding phenomena.The inspiration for the dynastic change strategy is that a dynasty can never be sustained forever in a society that changes continuously over time and with its environment.This approach solves the possible premature convergence problem.As a check of our method,we employed it to analyze 2 proteins 1AGT and 1AHO,and obtained energy minima of -20.8296 and -21.0853,respectively. These values are not only comparable to the reported literature values of -19.6169 and -15.1911 for these 2 proteins but are between 6-38%smaller than them.For these 2 proteins,we are reasonably confident that our simulation for the ground state conformation corresponds to the state of lowest free energy.In the second part of the thesis,we presented a technique for the prediction of the protein secondary structure based on a statistical method using short sequences of amino acids.This formulation is an extension of the traditional approach in which the secondary structure is predicted using a single amino acid.From extensive numerical experimentations to assess the optimum length of the short sequence,we found that the sequence involving 3,4,5,6 amino acids provides the best results.Adopting this sequence,we obtained Q3=89.9%,88.8%and 89.2%,and SOV=84.3%,82.4%and 84.1%for the following test groups of the 126,396 and 2180 protein sets,respectively. We then analyzed the new 153 protein set sourced from the 2007-11-15 Whole PDB database and compared its Q3 and SOV results predicted by our method against the results computed by 3 well established techniques of GORâ…£,GORâ…¤and Jpred.Our results of Q3=73.7%and SOV=68.2%compare favorably with the averaged values of Q3=69.7%and SOV=66.9%from the 3 techniques.We feel that our short amino acid sequence approach is a promising technique for protein secondary structure prediction that can handle large protein databases with acceptable accuracy.