| Summary: | Three-dimensional (3D) structure of a protein is essential as the guidance of structure-based drug dis-covery. To achieve robust homology modeling with atomic-level accuracy, reliable loop predictions are required. Here, a novel hierarchical protocol of Protein Local Optimization Program (PLOP) is designed to produce sub-2 angstrom predictions on loop regions in homology modeling. Dramatic improvements in both speed and accuracy have been realized with implementation of special-designed clustering and adaptive loop closure algorithm. Four prediction rounds are designed for homology modeling as the high-level protocol of PLOP, which allows latter rounds employ the educated guess of backbone atom positions and hydrogen bonding information inherited from the previous rounds, contributing to additional prediction accuracy. The success of PLOP has been demonstrated with four different data sets, mainly concen-trating on homology modeling of H3 loops of antibodies. GPU-accelerated sampling algorithm and deep learning models are implemented, which are able to produce promising predictions as input templates for PLOP in the context of homology modeling.
|
|---|
