Improving de novo protein structure prediction using contact maps information

Abstract

The use of residue-residue contact maps in protein structure prediction (PSP) has proved promising during the last CASP editions (CASP10, 11 and 12). The goals of this work are to carry out an assessment of the information given by contact maps and to develop a strategy to use the contact constraints from these maps to improve the quality of the predicted models in a de novo PSP approach. A residue-residue potential, with information from contact maps, is proposed in the form of distance constraints. This potential is added to the fitness function of the GAPF program, which predicts protein structures using a genetic algorithm with phenotypic crowding in a free-modeling approach. Two contact maps were generated to evaluate the potential developed here: (i) a native contact map obtained directly from the experimental structure and, (ii) a filtered contact map with only the native contacts present in a map predicted by MetaPSICOV. The experiments performed indicate that the contact potential implemented in the GAPF program promoted an important improvement in the accuracy of the predictions, confirming the use of contact maps as a useful strategy for de novo PSP methodologies. Our results also stress the need to develop better strategies to filter and enhance the information of predicted contacts.