We used the original graphical output of each web server to
prepare Fig.3. In this figure, the first row displays the query
sequence with highlighted experimentally known drug binding
sites. This is followed in successive rows by the results of the
functional site prediction methods explained in this chapter (i.e.,
Tracesuite II, Universal ET, ConSurf, and PROFisis). In Fig.3,we
only illustrated the residues between the positions 594 and
812 (in three blocks), since imatinib is known to bind to the
residues in this region. TraceSuite II gives the output on the
residues of the given structure by highlighting buried and con-
served residues in light gray. If the exposed and conserved residues
are of interest, they can be browsed at the resulting evolutionary
trace, which was mapped to the given structure. Universal Evolu-
tionary Trace method provided a histogram that was colored
according to the importance of each residue. ConSurf was used in
the sequence input mode (ConSeq tool) and provided conservation
scores along with exposed vs. buried and functional vs. structural
residue predictions. For PROFisis, binding sites, secondary struc-
tures, and solvent accessibilities are shown at different rows.
As shown in Fig.3, all methods recover most of the imatinib
interacting binding sites located on KIT protein, except PROFisis.
As observed, most of the methods gave high scores to the residues
around the actual binding sites, as well. There can be two possible
reasons behind this: (1) the residues at the proximity of binding
sites may also play roles in binding; as a result, they are conserved,
and (2) the predictive approaches often consider the properties of
the neighbors of the corresponding residue while calculating con-
servation/importance, which results in a smoothed scoring curve.
Apart from that, the tested methods also predicted a few additional
residues on the sequence, which may be targets for other molecules,
located in the core region or just false positive hits.4 Notes
- The parameters of BLAST are as follows: The first option is the
target database. In this menu, BLAST search database can be
selected as either UniProtKB (as a whole or divided to taxo-
nomic nodes, or including only reviewed sequences in
UniPRotKB/Swiss-Prot) or UniRef (i.e., sequence clusters at
different similarity levels, which is useful for eliminating redun-
dant sequences) or UniParc (i.e., sequence archive is the largest
set possible including the deleted sequences as well, not advised
to be used unless for specific objectives). The second parameter
option is the E-value threshold, which is a threshold statistical
measure indicating the number of returned matches. E-values
lower than 0.1 are generally accepted as significantly similar
hits. It is also possible to select a higher E-value to enlarge the
Phylogenetics-Based Prediction of Functional Sites 65