Investigations into the effect of multiobjectivization in protein structure prediction

Research output: Contribution to journalArticlepeer-review


Physics-based potential energy functions used in protein structure prediction are composed of several energy terms combined in a weighted sum. 'Multiobjectivization' - splitting up the energy function into its components and optimizing the components as a vector using multiobjective methods - may have beneficial effects for tackling these difficult problems. In this paper we investigate the hypotheses that multiobjectivization can (i) reduce the number of local optima in the landscapes, as seen by hillclimbers, and (ii) equalize the influence of different energy components that range over vastly different energy scales and hence usually swamp each other's search gradients. The investigations use models of two real molecules, the alanine dipeptide and Metenkephalin under the Amber99 energy function, and consider hillclimbers with a range of mutation step sizes. Our findings support the hypotheses and also indicate that multiobjectivization is competitive with alternative methods of escaping local optima. © 2008 Springer-Verlag Berlin Heidelberg.
Original languageEnglish
Pages (from-to)702-711
Number of pages9
JournalLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Publication statusPublished - 2008


Dive into the research topics of 'Investigations into the effect of multiobjectivization in protein structure prediction'. Together they form a unique fingerprint.

Cite this