Genetic operators based on backbone constraint angles for protein structure prediction

This paper describes the development of genetic operators for protein structure prediction by using the frequency of occurrence of the main chain dihedral angles observed in 5,387 experimentally determined protein structures. The use of constraints in the conformation space accelerates the search by the genetic algorithm and improves its predictive capacity. The operators were tested in a crowding-based steady-state genetic algorithm implemented in the GAPF program, using a test set of eight proteins belonging to different classes (i.e., preferably α, α/β and preferably β). The results show that using operators with restricted backbone dihedral angles ranges, for each amino acid and secondary structure type, reduces up to 75% the number of energy function evaluations required to obtain equivalent results to those obtained without the new operators. Furthermore, these new operators increased the capacity of the algorithm to obtain better protein models, since for the two largest proteins from the test set, i.e. 1BDD and 1GYZ, the best predicted models had their GDT-TS value increased by more than 10% comparing with those values obtained using the standard version of the algorithm.