A Comparative Study on Single and Multiple Point Crossovers in a Genetic Algorithm for Coarse Protein Modeling

The protein structure prediction problem is a holy grail for life science researchers. Computational protein structure prediction involves the folding of protein sequence (string) into the tertiary structure, called the native protein structure. The hydrophobic polar (HP) model is one of the basic models used to investigate the protein folding mechanism at the coarse level. In the HP model, the protein structure prediction problem is defined as an optimization problem, where the protein sequence must be folded over a lattice space such that the protein structure exhibits the lowest value of free energy. However, with the HP model, protein structure prediction is a nondeterministic polynomial (NP)-complete problem and is, therefore, simulated using meta-heuristic algorithms. Simulation of the HP model results in the formation of various protein structures called protein conformations. In this article, we present a case study on the application of a genetic algorithm to simulate the HP model based protein structure prediction. In this work, we employ the two versions of crossover functions (single-point vs. multiple-point crossovers) to generate protein conformations. The conformations were assessed based on the presence of hydrophobic contacts identified in the experimental structure. The sensitivity, specificity, and accuracy of simulation algorithm (genetic algorithm) were compared, and the significance of the parameters was statistically evaluated using the paired t-test. Our results indicate that the multipoint crossover operator enhanced the performance of genetic algorithm compared to genetic algorithm with single-point crossover. Also, multipoint crossover reduced the generation of false conformations, which results in a significant reduction in computational cost.