Protein structure prediction using diversity controlled self-adaptive differential evolution with local search

In this paper, Protein Structure Prediction problem is solved using Diversity Controlled Self-Adaptive Differential Evolution with Local search (DCSaDE-LS). DCSaDE-LS, an improved version of Self-Adaptive Differential Evolution (SaDE), use simple fuzzy system to control the diversity of individuals and local search to maintain a balance between exploration and exploitation. DCSaDE-LS with four different local search replacement strategies are used. SaDE is also implemented for comparison purposes. Algorithms are tested on a peptide Met-enkephalin for force fields ECEPP/2, ECEPP/3 and CHARMM22. Results show that both DCSaDE-LS and SaDE produce the best energy for both force fields. Among the four replacement strategies, DCSaDE-LS1 strategy reports better results than other strategies and SaDE in terms of number of function evaluations, mean energy and success rate. Best conformations obtained using DCSaDE-LS is compared with native structure 1PLW and GEM structure Scheraga. Nonparametric statistical tests for multiple comparisons (\(1\times N\)) with control method are implemented for CHARMM22 observations. A set of unique 100 best conformations obtained from DCSaDE-LS are clustered into 3 independent clusters suggesting the robustness of this methodology and the ability to explore the conformational space available and to populate the near native conformations.