Many-objective approach based on problem-aware mutation operators for protein encoding

Abstract

One of the research pathways in synthetic biology is protein encoding, which aims to improve and increase protein expression. The production of heterologous proteins is important in different fields, such as medicine, environmental biology, and agriculture. Protein encoding is a difficult task and can be defined as a many-objective optimization problem, where four objectives must be optimized concurrently: codon adaptation, guanine-cytosine content, difference between sequences, and avoidance of hairpin loops. These objectives are why we propose and describe a many-objective approach based on the non-dominated sorting genetic algorithm-III (NSGA-III) for protein encoding. The proposed algorithm was designed to work with novel mutation operators that are problem-aware and consider the different optimization objectives. The results of the proposed algorithm were compared to those of different tools that were reported by other authors to determine its efficiency. Comparisons were made using several quality metrics and with statistical analyses. The average improvements in hypervolume and set coverage were between 5.34% and 55.75%, and 53.01% and 100%, respectively. Based on the statistical analyses, the proposed algorithm was found to produce statistically significant improvements compared to other algorithms, thus highlighting the relevance of the proposed approach for protein encoding.