Comparative Analysis of Techniques Oriented on the Recognition of Ligand Binding Area in Proteins - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 60

In our calculations we have applied MUSCLE, as the authors claim that it is both
more efficient and more accurate than the alternative solution. Subsequently,
the tool constructs a phylogenetic tree by applying the neighbour joining (NJ)
algorithm (Pupko et al. 2002 ). The resulting tree consists of homologues singled out
in the previous stage. Finally, a sequence of “conservation scores” is calculated
using empirical Bayesian (Mayrose et al. 2004 ) or Maximum Likelihood (Pupko
et al. 2002 ) algorithms. In our research we applied the default “Evolutionary
Substitution” settings.
The end result is a three-dimensional representation of the protein structure,
which can be visualized using FirstGlance (Ashkenazy et al. 2010 ) . The surface
of the protein is tagged with “conservativeness scores” using various colors. ConSurf
also generates output pdb files, containing the structure of the protein along
with the identified amino acids for which “conservativeness scores” have been
determined.

4.4.2.2 Fuzzy Oil Drop Model

This model relies on identifying irregularities in the distribution of hydrophobicity
within the protein molecule. These irregularities are then compared with an ideal-
ized, theoretical distribution obtained by using a 3D Gauss function (Konieczny
et al. 2006 ; Banach et al. 2012 ) as representing the highest hydropgobicity at the
central part of ellipsoid (or sphere if the size of drop is equal along each direction)
with hydrophobicity decrease according to the increase of distance versus the center
of the protein molecule reaching the level close to zero at the surface of protein
body. Such idealized hydrophobicity distribution is expected to be identified in a
special group of proteins (like downhill proteins). It is assumed that the irregularity
of hydrophobicity distribution in protein body represents the intentional character
what means it is function related. Thus comparison of expected and theoretical H
pro fi les is performed. Plotting the distribution of

Δ H (differences between expected
and observed hydrophobicity) along the polypeptide chain reveals residues for
which

Δ H reaches high values on the positive or negative scale. According to
the theoretical model, the former are suspected of involvement in binding ligands
(usually of the hydrophobic or emphiphilic variety) while the latter - if exposed on
the surface of the protein - may participate in protein complexation, resulting in
multi-protein aggregates.
Ligand binding sites are thus determined by searching for residues with either
very high or very low

Δ H values (local minima or maxima). Proper identification
relies on selecting a cutoff threshold, starting with the

Δ H pro fi le maximum.
Performing calculations for consecutive cutoff thresholds yields a ROC curve which
represents the given protein. Accuracy of predictions may then be determined
by calculating the surface area bounded by two curves - a diagonal and the TPR-
vs-FPR curve. The greater the area, the more accurate the results.