Bioinformatics Analysis to Identify Cell Adhesion Molecules in Cancer - Adhesion Molecules - page 316

sequence database searches locally and is freely available online (see Table 1 )
(Sadanandam et al . 2008a). Using this approach, we identii ed less than 25 DNA
or protein hits for each phage-display peptide selected in vivo , whereas using non-
redundant databases of all proteins resulted in more than 150 hits (Sadanandam
et al . 2007).
As a robust approach to i nding tumor-associated CAMs, we developed a
new motif search program called Similar enRiched Parikh Vector Searching
(SRPVS, see Table 1 ) (Huang et al . 2005) by assuming that when two sequences
are homologous, they share biological structure, irrespective of sequence order.
In this method, a protein sequence under analysis is split into a pre-dei ned word
size and its sequence similarity to another protein is detected in a manner that
is l exible with respect to local sequence order. h e order dif erence between the
two sequences is scored. h is method will also analyze proteins with reverse or
shul ed order of their ligand-binding sequences (Huang et al . 2005). In addition,
the RELIC sot ware (described above) applies the bioinformatics summation of
multiple binding sequences to produce collective ligand-binding signatures that
can be used to search for the tumor-associated proteins (Mandava et al . 2004).
Structure and Gene Expression of Tumor-associated Proteins
Once the putative tumor-associated proteins are identii ed using the preceding
step, the next step is to identify the structural signii cance of the peptide sequences
at the ligand-binding regions of tumor-associated CAMs and then to screen the
gene expression of tumor-associated CAMs to understand their role in cancer
(Fig. 3) . To our knowledge, not many phage-display-based bioinformatics analyses
have considered these two steps. But it is essential to understand that without these
two steps the whole process of genome-based identii cation of biomarkers is not
complete. In order to coni rm that the identii ed phage-display peptides are part
of the ligand-binding region, we successfully used the protein modeling server
CPHModels (Table 1) to perform three-dimensional molecular modeling based
on the known or predicted structures of the tumor-associated CAMs (Fig. 3)
(Sadanandam et al . 2007). RELIC sot ware also has a set of programs that allows
the user to compare peptides to sequence of known structures (Mandava et al .
2004).
To analyze gene expression of putative tumor-associated binding proteins in
tumor tissues, gene expression databases such as gene expression atlas, Serial
Analysis of Gene Expression data repository (SAGEmap) and TisssueInfo
(database of tissue-specii c gene expression) can be used (Boon et al . 2002, Scherf
et al . 2000) (Fig. 3) . h e current version of the MCAM 3.0 database is an interactive
Web-based database and provides the resources needed to search function and
gene expression in normal and tumor tissues for a given gene (Sadanandam et al .
2008a). We have performed such analysis and have identii ed 30 putative CAMs.