Authors:
Reitze Jansen
1
;
Ruben Horn
2
;
3
;
Okke van Eck
3
;
Kristian Verduin
1
;
Sarah Thomson
4
and
Daan van den Berg
1
Affiliations:
1
Department of Computer Science, University of Amsterdam, Netherlands
;
2
Helmut-Schmidt-University, Hamburg, Germany
;
3
Department of Computer Science, VU Amsterdam, Netherlands
;
4
Napier University, Edinburgh, U.K.
Keyword(s):
Protein Folding, Genetic Algorithms, Evolutionary Computing, Constraints, Constraint Hierarchy.
Abstract:
Genetic algorithms might not be able to solve the HP-protein folding problem because creating random individuals for an initial population is very hard, if not impossible. The reason for this, is that the expected number of constraint violations increases with instance size when randomly sampling individuals, as we will show in an experiment. Thereby, the probability of randomly sampling a valid individual decreases exponentially with instance size. This immediately prohibits resampling, and repair mechanisms might also be non-applicable. Backtracking could generate a valid random individual, but it runs in exponential time, and is therefore also unsuitable. No wonder that previous approaches do not report how (often) random samples are created, and only address small instances. We contrast our findings with TSP, which is also NP-hard, but does not have these problems.