We are pursuing concepts from multi-objective optimization in order to attenuate the reliance of energy-guided probabilistic search algorithms on noisy energy functions. The objective is to adress the fact that all energy functions that sum independent terms contain inherent errors and often guide search towards states that are artifacts of a given energy function constructed on a specific set of structures. The key idea is to group terms of an energy function into different categories and drive the exporation towards conformations that optimize these categories, effectively reproducing a more diverse set of physically-realistic structures.

We have pursued several Pareto-based metrics in the context of hybrid (memetic) population-based EAs and shown their superiority in the context of decoy sampling for the problem of ab-initio structure prediction. This work has appeared in: 1) Brian Olson and Amarda Shehu. "An Evolutionary-inspired Algorithm to Guide Stochastic Search for Near-native Protein Conformations with Multiobjective Analysis." AAAI Workshop on Artificial Intelligence and Robotics Methods in Computational Biology, Bellevue, Washington, 2013. Further Pareto-based metrics and comparisons with the popular Rosetta framework has been recently published in: Brian Olson and Amarda Shehu. "Multi-Objective Stochastic Search for Sampling Local Minima in the Protein Energy Surface. ACM Bioinformatics and Computational Biology (BCB), Washington, D.C., 2013. A journal article on this line of investigation will appear soon.