Software:Model Quality Assessment General Description
From Honiglab_public
Global and Local Quality Assessment of Protein Models
Three protocols have been implemented for model quality assessment (MQA), including direct assessment of local qualities using statistical potentials and two machine-learning based protocols.
The statistical potentials used here include newly developed solvation potentials, environmental potentials, pairwise potentials, hydrogen bonding potentials and backbone dihedral potentials. The normalized version of these potentials can be directly used to evaluate local qualities by plotting against residue numbers.
The two machine-learning based protocols combine a number of state-of-the-art statistical potentials and scoring functions using support vector machines (SVMs). The scoring functions used here include two tabulated physical functions that were derived from the OPLS united-atom force field and three functions that primarily compare secondary structure assignment by DSSP and secondary structure prediction by PSIPRED. One protocol is aimed to predict global quality of protein models measured by C-alpha RMSD (CARMS) of backbone, C-alpha RMSD of secondary structure elements (SSRMS), GDT_TS, MaxSub and TM-score. The other protocol is designed to predict the so-called S-score proposed by Wallner and Elofsson (Protein Sci. 2006). A window of 11 residues is used to evaluate the local quality of the center residue. A large number of models have been compiled from CASP5, CASP6 and CASP7 for the SVM training and testing. More methodological details can be found in the upcoming paper by Zhu and Honig (in preparation).
Model Quality Assessment Program is supported by a funding from the NIH Grant # GM30518
Developed in the Honig Lab
