Hybridized distance- and contact- based hierarchical protein structure modeling using DConStruct

Crystallography and NMR system (CNS) is a widely used method for predicting 3D structures of protein from inter-residue distance or contact maps. However, the decade-old CNS is an experimental structure determination method that was originally developed for solving macromolecular geometry from experimental restraints, as opposed to predictive structure modeling. Thus, relying on CNS for structure modeling may undermine the ab initio folding performance. Here we propose a CNS-free protein structure modeling method called DConStruct [1], which performs 3-stage hierarchical predictive modeling with iterative self-correction driven purely by the geometric restraints induced by inter-residue interactions and secondary structures. Starting from a residue-residue interaction map and secondary structure, DConStruct can hierarchically estimate the correct overall fold of a target protein in coarse-grained mode to progressively optimize local and non-local interactions while enhancing the secondary structure topology in a self-correcting manner. Multiple large-scale benchmarking experiments show that our proposed method can substantially improve the folding accuracy for both soluble and membrane proteins compared to state-of-the-art approaches. The open-source DConStruct software package, licensed under the GNU General Public License v3, is freely available at https://github.com/Bhattacharya-Lab/DConStruct.