Differential Evolution Multi-Objective for Tertiary Protein Structure Prediction

Abstract

The determination of proteins’ structure is very expensive and time-consuming, making computer-aided methods attractive. However, in computational terms, the protein structure prediction is a NP-Hard problem [17], meaning that there is no efficient algorithm that can find a solution in a viable computational time. Nonetheless, the energy terms that compose different force fields seem to be conflicting among themselves, leading to a multi-objective problem. In this sense, different works in the literature have proposed multi-objective formulations of search mechanisms. Hence, we use the Differential Evolution Multi-Objective (DEMO) algorithm with the Rosetta score3 energy function as a force field. In our work, we split the energy terms into two objectives, one with only the van der Waals values, while the second one contains the remaining bonded and non-bonded, including the secondary structure reinforcement. Moreover, we enhance the DEMO algorithm with structural knowledge provided by the Angle Probability List (APL). From this perspective, our work provides different contributions to the research area, since the DEMO algorithm was never used in this problem, neither the APL with this algorithm. Also, the multi-objective formulation using Rosetta score3 was not yet explored by related works, even though its relevance for the problem. Results obtained show that the DEMO found better structures than the single-objective differential evolution that uses the same mutation mechanism, energy function, and APL. Also, DEMO reached competitive results when comparing with state-of-art bi-objective approaches.