Home - Comparative modelling - Contact - Secondary structure - Prev - Next - Bottom

EVA measures for threading prediction accuracy

Author: Dariusz Przybylski dudek@cubic.bioc.columbia.edu

Contents



    definitions of accuarcy measures


The following scores are used to measure threading prediction accuracy
note: most tables contain only a subset of these scores):
  1. P_MMT
    P-value of structural alignemnt of medel and
    experimental structures assigned by program
    MAMMOTH
  2. Z_CE
    Z-score of structural alignment of model and
    experimental structures assigned by program CE
  3. Q_LG1
    quality score for structural alignemnt of medel and
    experimental structures assigned by program LG1
  4. Q_LG2
    quality score for structural alignemnt of medel and
    experimental structures assigned by program LG1
  5. NXA_Gbl%
    percentage of equivalent residues within a distance
    of X Angstroms in a global superposition of model
    and experimental structures
  6. NXA_ext_CE%
    percentage of target residues found within a distance
    of X Angstroms in structural alignment of model and
    target structures seeded by CE and extended iteratively
  7. ModN%
    percent of target residues in model produced by method
  8. pairwise score
    Let Nb - number of times server is signinficantly better
    than other servers.
    Let Nw - number of times server is signinficantly worse
    than other servers.
    Let N -number of times a comparison can be made.
    Than: score = 10 * ( Nb - Nw )/N .
    The significance is computed using Student t-distribution
  9. ROC curves
    plots of true positives vs false positives for prediction
    sets ordered according to internal score of a method.
    Cutoff values for true and false positives are defined
    by "Sgnf(measure)" values above
  10. CE classification
    Classifcation into Family, Superfamily and Twilight
    zone is based on z-score from CE statistical model.
    >4.5- family level similarity;
    4.0-4.5 - superfamily level similarities, strong
    function related similarities or strong
    recurring fold;
    3.7-4.0 - twilight zone where some similarities
    of biological significance can be seen;
    <3.7 - similarities of low significance, but still
    some biologically important similarities
    can be revealed, but interpretation normally
    requires additional evidence.
    We take scores above 3.7 as significant



    References







Home - Comparative modelling - Contact - Secondary structure - Prev - Next - Top