Multi-Objective Evolutionary Algorithm for Discovering Peptide Binding Motifs
Multi-Objective Evolutionary Algorithms (MOEA) use Genetic Algorithms (GA) to find a set of potential solutions, which are reached by compromising trade-offs between the multiple objectives. This paper presents a novel approach using MOEA to search for a motif which can unravel rules governing peptide binding to medically important receptors with applications to drugs and vaccines target discovery. However, the degeneracy of motifs due to the varying physicochemical properties at the binding sites across large number of active peptides poses a challenge for the detection of motifs of specific molecules such as MHC Class II molecule I-Ag7 of the non-obese diabetic (NOD) mouse. Several motifs have been experimentally derived for I-Ag7 molecule, but they differ from each other significantly. We have formulated the problem of finding a consensus motif for I-Ag7by using MOEA as an outcome that satisfies two objectives: extract prior information by minimizing the distance between the experimentally derived motifs and the resulting matrix by MOEA; minimize the overall number of false positives and negatives resulting by using the putative MOEA-derived motif. The MOEA results in a Pareto optimal set of motifs from which the best motif is chosen by the Area under the Receiver Operator Characteristics (AROC) performance on an independent test dataset. We compared the MOEA-derived motif with the experimentally derived motifs and motifs derived by computational techniques such as MEME, RANKPEP, and Gibbs Motif Sampler. The overall predictive performance of the MOEA derived motif is comparable or better than the experimentally derived motifs and is better than the computationally derived motifs.
KeywordsPeptide Binding Consensus Motif Multiobjective Evolutionary Algorithm Experimental Motif Anchor Position
Unable to display preview. Download preview PDF.
- 4.Fonseca, C.M., et al.: Genetic Algorithms for Multiobjective Optimization: Formulation, discussion and generalization. In: Forrest, S. (ed.) Proc. of the fifth Intl. conference on Genetic Algorithms, pp. 416–423. Morgan Kauffman, San Mateo (1993)Google Scholar
- 5.Lee, S., et al.: Comparison of Multi-Objective Genetic Algorithms in Optimizing Q-Law- Thrust Orbit Transfers. In: Lee, S., et al. (eds.) GECCO (2005)Google Scholar
- 6.Amor, S., et al.: Encephalitogenic epitopes of myelin basic protein, proteolipid proteing, and myelin oligodendrocyte glycoprotein for experimental allergic en-cephalomyelitis induction in Biozzi AB/H(H-2Ag 7) mice share an amino acid motif. J. Immunology 156, 3000–3008 (1996)Google Scholar
- 7.Reich, E.P., et al.: Self peptides isolated from MHC glycoproteins of non-obese diabetic mice. J. Immunology 152, 2279–2288 (1994)Google Scholar
- 14.Yu, B., et al.: Binding of conserved islet peptides by human and murine MHC class II molecules associated with susceptibility to type I diabetes. J. Immunology 30(9), 2497–2506Google Scholar
- 16.Stratman, T., et al.: The I-Ag 7 MHC class II molecule linked to murine diabetes in a promiscuous peptide binder. J. Immunology 165, 3214–3225 (2000)Google Scholar
- 17.Brusic, V.: An unpublished datasetGoogle Scholar
- 19.Bailey, T.L., Elkan, C.: The value of prior knowledge in discovering motifs with MEME. In: Proc. Int. Conf. Intell. Syst. Mol. Biol., vol. 3, pp. 21–29 (1995)Google Scholar
- 24.Coello, C.A.: An Updated Survey of Evolutionary Multiobjective Optimization Techniques: State of the Art and Future Trends. In: Congress on Evolutionary Computation, pp. 3–13. IEEE Service Center, Washington (1999)Google Scholar