A knowledge-based approach to initial population generation in evolutionary algorithms: Application to the protein structure prediction problem

Abstract

In this study we introduce a new approach to generate the initial population of an Evolutionary Algorithm (EA), based on problem-specific knowledge. We discuss the key ingredients (knowledge and diversity) necessary to generate a good diverse initial random population, with particular application to the protein structure prediction problem. Two main components of our Initial Population Generation (IPG) algorithm are described: (a) one provides the bio-chemical problem-specific knowledge (Molecular Dynamics (MD) and Normal Mode Analysis (NMA)); (b) the second one is an algorithm which ensures population diversity by using the complete graph of the generated bio-molecular conformations. Results show that IPG is a promising algorithm for the creation of good diversity initial populations.