Biology Reference
In-Depth Information
4.4.2.3
SuMo (Sur fi ng the Molecules)
The model implemented by this program is based on the analysis of 20,000
proteins with known 3D structures. 11,000 categories of ligand binding sites have
been recognized by the authors (Jambon et al.
2003,
2005
). The protein under con-
sideration is compared against all categories. The geometry of the binding site is
represented as a simplified graph, based on chemical group triplets. The positions of
groups with respect to one another (edge lengths) determine each functional trian-
gle. Graph similarity is taken as an identification criterion, where the resulting graph
is compared with the 11,000 previously mentioned categories. Potential protein-
ligand interaction sites (as well as the recognized ligand molecules) are produced as
output, along with a ranking list of all possible solutions.
The analysis described in this work bases on the following user-defined options:
-
single selected chain;
-
single, specific ligand (FMN or NAD
+
);
amino acid numbers extracted from a text-based output file written by the soft-
-
ware tool;
TP, FP, FN and TN values calculated by relying on PDBSum data (Laskowski
-
2009
) .
4.5
Ligand Binding Site Recognition - Comparable Analysis
Results obtained with CASTp, Pocket-Finder, QSite-Finder, ConSurf, Sumo and
the Fuzzy Oil Drop (FOD) model were subjected to validation, as described above.
Rankings obtained by applying the MCC criterion diverge from corresponding
F-measure rankings - in particular, the top and bottom parts of each list tend to
contain the same proteins, but in a different order. Proteins for which MCC and
F-measure values could not be calculated, or for which the programs did not arrive
at any solutions, were omitted. We also skipped proteins which could not be correctly
processed by a given program due to size restrictions - thus, a different number of
solution is listed for each software package.
Results are presented separately for NAD
+
and FMN ligands and subdivided into
geometry- and knowledge-based models. Within each group software packages
are discussed alphabetically.
4.5.1
NAD
+
Complexing Proteins
4.5.1.1
Geometry-Based Packages
Figure
4.1
presents the results of geometry-based analysis of NAD
+
binding pockets,
as performed by the CASTp tool. The MCC measure exhibited the lowest variance
of binding site identification validity (compared to other software packages),
Search WWH ::
Custom Search