Robust Bio-active Peptide Prediction Using Multi-objective Optimization

Bio-active peptides control many important functions in organisms, such as cell reproduction, appetite, euphoria, sleep, learning, immune response, etc. They also act on hormones, neurotransmitters, antioxidants, toxins and antibiotics. Because of their importance, bioactive peptides have received particular attention and extensive studies have been carried out to determine their structures. Although their typical size does not exceed 30 amino acids, their 3D structure is challenging to predict because of the following reasons: (i) their conformation often includes beta-turns which are more difficult to predict using standard potential energy functions; (ii) they fold into structures that are not similar to already known proteins, which makes them hard instances for comparative modeling techniques; (iii) they are more exposed to the solvent than longer proteins and this additional effect has a consequence on their final conformation. This paper presents a strategy for peptides structure prediction that uses: (1) a multi-objective formulation of the optimization problem, (2) a multi-objective evolutionary algorithm to explore the search space, (3) a decision making phase based on different metrics to select solution from the Pareto front, and (4) a method to analyze the robustness of the solution using the Monte Carlo method. We have tested this prediction pipeline on a large dataset of 43 bioactive peptides and the experimental results show that this method outperforms the PEPstr prediction server and is competitive against a more recent Generalized Pattern Search approach. Multiple solutions can be generated, as opposed to standard single-objective methods, which are generally more robust than the wild-type.