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-A^g7 of the non-obese diabetic (NOD) mouse. Several motifs have been experimentally derived for I-A^g7 molecule, but they differ from each other significantly. We have formulated the problem of finding a consensus motif for I-A^g7 by 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 (A_ROC) 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.