Match-Box Web Server 1.3

Guy Baudoux, Christophe Lambert, Ernest Feytmans and Eric Depiereux.

Molecular Biology Research Unit, The University of Namur, Belgium.

Project supported by the Walloon Government, Federal State of Belgium.

Return to SUBMIT | HELP page.


New features.

1) Cutoffs have been refined in order to fit better 33 test cases of known structure and low level of identity.

If you compare the outputs of the current and of the previous version, you should observe only slight modifications in the boxes delineation (border effects) on reliable alignments. However, for alignments including unrelated sequences, the boxes of low reliability may be substantially different, because the filtering of random noise is not exactly the same.

2) The scale of reliability scores has been revisited.

A score from 1 to 9 is written below each position in the boxes. It corresponds to the level of similarity at which this position has been included in the box and is thus related to the statistical significance of the alignment at this position. Lower the score is, higher is the reliability of the alignment. A score of 5 corresponds to a level of similarity of equal occurrence in related and unrelated sequences. This score is related to the averaged OBSERVED confidence rate by a quite perfect linear relationship. For scores below 5 on more than 2 non redundant sequences, you can expect a rate of confidence over 90% (less than 10% of false positive), even when the percentage of identity is very low (10%). The following results (Table 1) have been obtained on 20 families of 3 to 6 known structures sharing between 9% and 44% of conserved residues.

Table 1. Percentage of correctly predicted aligned residues, with respect to structure alignments. Results expressed in % confidence (correctly predicted/total aligned) Minimum and maximum observed on the 20 tests.
Reliability ScoreMinimum %Maximum %
641.386.8
548.8100
473.9100
3 84.6100
2100100

Return to SUBMIT | HELP page.


Webmaster