An Evolutionary Model Based on Hill-Climbing Search Operators for Protein Structure Prediction

A hybrid evolutionary approach to protein structure prediction with lattice models

“…This results in the generation of − 1 new individuals from which the one having the best fitness is selected. Similar to the hill-climbing strategy for recombination, the hill-climbing mutation in ELF differs from the one for EAHCD [2] as a mutated individual is not subsequently engaged in further hill-climbing iterations. This new more "weak" approach of integrating hill-climbing mechanisms into the search process is able to trigger good results while clearly saving computational time.…”

“…Only the best fitted offspring (out of the individuals created) is selected and used in replacement strategies. This is a different and less computationally expensive hill-climbing crossover compared to the strategy presented in [2] where offspring are generated from pairs of parents within a predefined number of hill-climbing iterations. Uniform crossover is slightly customized to allow the generation of valid configurations: whenever a value from one parent can not be accepted due to collisions, a valid direction randomly selected is chosen.…”

“…The main scheme of the proposed ELF model is inspired by the EAHCD algorithm for PSP described in [2]: each individual represents a protein configuration, the population is randomly initialized with valid conformations, crossover and mutation are applied using a hill-climbing mechanism and diversification is reinforced periodically. The algorithm uses uniform crossover adjusted for the recombination of protein configurations.…”

“…The currently proposed model uses uniform crossover and the same pull-move transformations which are engaged in a customized and more simple hillclimbing mechanism able to save computation time while preserving the quality of the search process (this feature is fully detailed in the third section of the paper). A major improvement of the ELF algorithm compared to the model presented in [2] refers to the diversification strategy which becomes more meaningful by comparing the fingerprint of the protein conformations and not just the individual values (to decide how similar are two given individuals). This new fingerprint-based diversification further affects the selection process particularly for the individuals to be replaced with newly generated genetic material.…”

“…The proposed model is called -evolutionary algorithm hybridized with local hill-climbing search and fingerprint-diversification. The starting point of ELF is formed by the evolutionary algorithm based on hillclimbing search operators ( ) introduced in [2]. The EAHCD algorithm uses a one-point crossover and pull-moves (as the mutation operator) both applied in a steepest ascent hill-climbing manner.…”