1. How was the local accuracy estimated?
    The local accuracy was defined as the distance deviation (in Angstrom) between residue positions 
    in the model and the native structure. It was estimated using support vector regression that makes 
    use of the coverage of threading alignment, divergence of structural decoys, and sequence-based 
    secondary structure and solvent accessibility predictions. 
    Large-scale benchmark tests show that we are able to predict the local accuracy with an average 
    error of 2.21 Angstrom and Pearson's correlation coefficient 0.7. 

    The threading alignment coverage, predicted secondary structure and solvent accessibility are 
    visualized to help users make better use of the estimated local accuracy of models. 
    Based on our tests, the local accuracy estimations tend to be  more accurate for residues:
	 1) that have higher threading alignment coverage
	 2) that are located at alpha-helix and beta-strand regions
	 3) that are buried (at 25% threshold)

    The estimated local accuracy for each model is available at the columns 61-66 in the model's PDB file
    and at the end this file.

2. What is B-factor profile?
    B-factor profile (BFP) for a target protein is defined as z-score-based normalization of raw B-factor values. 
    The BFP is predicted using a combination of both template-based assignment and profile-based prediction.
    Based on the distributions and predictions of the BFP, residues with BFP values higher than 2 are 
    less stable in experimental structures.

    The predicted BFP is available at the end of this file.


For more information about the local accuracy and B-factor profile predictions, please refer to the following article:

Jianyi Yang and Yang zhang, Predict residue-specific quality of I-TASSER models and B-factor profile, 2014 (submitted).