Biotech2012spring 1-overview 0

1359-6446/04/$ – see front matter ©2004 Elsevier Ltd. All rights reserved. PII: S1359-6446(04)03224-6
Cancer is a disease characterized by exten-
sive genomic abnormalities and aberrations
in gene expression. Tumor-specific mutations,
DNA amplifications and translocations can all
distort the normal program of gene expres-
sion and function, resulting in misregulated
activity of apoptosis, angiogenesis and cell
proliferation. A class of new cancer therapies,
including Avastin, Erbitux, Gleevec and
Herceptin, exploit our knowledge of cancer
biology by selectively inhibiting or killing
cancer cells, while leaving normal cells un-
harmed. Most of these novel therapies specif-
ically target cancer-causing gene products,
such as Bcr–Abl, or proteins required for tumor
progression and metastasis, such as vascular
endothelial growth factor (VEGF). Other can-
cer targets, such as CD20 for Rituxan, are bio-
markers and their physical presence confers
vulnerability to cancer cells, although it is not
clear whether they are causally implicated in
oncogenesis.
Identification of additional cancer targets
requires understanding the genetic events
associated with cancer genesis and progression,
as well as the identification of genes or gene
mutations that either promote oncogenesis or
are specifically associated with cancer. With
the advent of genomic technologies and their
application to the study of cancer biology, we
are presented with exciting opportunities to
perform HTS for novel cancer targets (Figure
1). Here we discuss the uses of bioinformatics
methods and data sources in finding poten-
tial new cancer drug targets.
Sequence analysis as a broad-based
selection method
Although a wide variety of protein classes are
involved in cancer, targets of interest fall into
three general categories: secreted proteins, cell
surface receptors and markers, and intracellu-
lar kinases. Secreted and cell-surface proteins,
such as VEGF and epidermal growth factor re-
ceptor (EGFR), are essential for intercellular
communication, in particular signaling of cell
proliferation and angiogenesis, and they are
physically accessible to monoclonal antibod-
ies, which have proven to be effective in treat-
ing cancers [1]. By contrast, kinases are pivotal
players in signal transduction and provide
assayable enzymatic activities for screening
small-molecule inhibitors. A recent survey of
human cancer genes indicated an over-repre-
sentation of protein kinases in a collection of
genes that are causally implicated in cancer [2].
Homologues of proteins involved in cancer-
related pathways might have related biologi-
cal roles. With the human genome completely
sequenced, and the availability of pair-wise
sequence alignment programs such as BLAST
[3] or hidden Markov model-based programs
such as the HMMER package [4], finding se-
quence homologues of known cancer genes
has become relatively straightforward. However,
new cancer targets might not bear sequence
similarity to known cancer genes. It is there-
fore necessary to cast a wider net to include
molecules with more general structural features
Bioinformatics and cancer
target discovery
Brian Desany and Zemin Zhang
Brian Desany
Zemin Zhang*
Department of Bioinformatics
Genentech
1 DNA Way, M.S. 93
South San Francisco
CA 94080, USA
*email: zemin@gene.com
reviewsresearch focus
795
DDT Vol. 9, No. 18 September 2004
The convergence of genomic technologies and the development of drugs
designed against specific molecular targets provides many opportunities
for using bioinformatics to bridge the gap between biological knowledge
and clinical therapy. Identifying genes that have properties similar to
known targets is conceptually straightforward. Additionally, genes can be
linked to cancer via recurrent genomic or genetic abnormalities. Finally,
by integrating large and disparate datasets, gene-level distinctions can
be made between the different biological states that the data represents.
These bioinformatics approaches and their associated methodologies,
which can be applied across a range of technologies, facilitate the rapid
identification of new target leads for further experimental validation.
www.drugdiscoverytoday.com
w

of existing cancer targets, such as the presence of a signal
peptide, a transmembrane domain or a protein kinase do-
main [5]. This can be achieved using a variety of computa-
tional prediction tools, including SignalP for signal pep-
tides [6] and TMHMM for transmembrane domains [7].
The number of genes found in this way is expected to be
large, and a majority of these genes might not be directly
associated with cancer. One way to trim down the large
pool of initial selection is to use biological assays to iden-
tify cancer-associated genes. This experimentally driven
process is illustrated by the successful identification of EG-
VEGF, which was initially found by sequence analysis to be
a novel secreted protein with no predictable function and
was subsequently shown to possess strong mitogenic activ-
ity in a cell-based assay [8]. A more general, efficient and
powerful approach is to use the bioinformatics methods
described below to maximize our functional knowledge by
exploiting high-throughput genomic
data sources, such as gene expression
assays and mutational studies, reduc-
ing the need for a large prior experi-
mental investment.
Gene expression profiling using
microarrays
Many techniques have been developed
for using microarray gene expression
data to study various facets of cancer
biology. Here we address some of
the more common computational ap-
proaches that have found applications in
target discovery and refer the reader to
a recent review [9] for a general overview
of microarray technology, experimen-
tal design and data processing issues.
Microarray analytical approaches
generally fall into two categories: su-
pervised and unsupervised learning.
Supervised learning uses the known
identities of samples and the expres-
sion patterns of a limited number of
genes to generate predictors that
are used to classify unknown samples
for diagnostic purposes [10,11].
Unsupervised learning involves using
a distance metric to group samples or
genes together into related groups, or
clusters [12], and has direct applica-
tions to target discovery. Candidates
for further validation as potential
targets can be defined by gene clusters
that have desirable expression characteristics, for example,
more highly expressed in tumors than in normal cells.
The most straightforward inference of differential ex-
pression can be made from a fold-change measurement,
which is the ratio of the expression levels of a given gene
between two samples. Interesting gene clusters would
therefore contain genes with a high fold-change between
tumor and normal cells. Although useful and intuitive,
fold-change suffers by not taking measurement noise or
biological variability into account, particularly because
cancer is inherently a heterogeneous disease. It has become
common to include as many tumor samples as possible in
a given analysis so that statistical methods, such as ANOVA
and t-like tests [13], can be used on a gene-by-gene basis to
compare groups of large numbers of samples to each other,
rank gene clusters and assign statistical significance to the
results.
796
DDT Vol. 9, No. 18 September 2004reviews research focus
Figure 1. Bioinformatics integration of heterogeneous biological data provides opportunities
for understanding cancer genetics and identifying therapeutic targets. Cancer cells are
genetically different from normal cells due to somatic point mutations and aberrations at
the genomic level (translocations, deletions and amplifications). The phenotypic
consequences of these genetic changes include overexpression of cell-surface markers
and activated kinases, presenting cancer-specific targets for therapeutic treatment. Various
types of data from expressed sequence tag (EST), microarray and comparative genomic
hybridization (CGH) technologies can be combined to detect point mutations, DNA copy
number changes and differentially expressed genes in cancer cells, leading to the identification
of cancer-causing genes as well as other genetic properties associated with cancer.
Drug Discovery Today
Gene/protein
Tissue/disease
ESTsSNPs
Genomic sequences
Microarray
CGH/array CGH
Available data
Bioinformatics
Integration
Normal cell
Somatic mutations
Genomic instability
(translocations, deletions and amplifications)
Differentially expressed genes
Distorted protein activities
Altered intracellular signaling
Cancer cell
Targets
Growth factors (e.g. VEGF)
Signal transducers (e.g. BRAF)
Cell surface markers (e.g. HER2)

Biological variability is partially mitigated by increas-
ingly precise sample acquisition techniques, such as laser
capture microdissection, that can reduce the contribution
of non-relevant cell types to the expression profiles.
However, cancers with similar histopathological features,
and even the same diagnosis, can actually represent more
than a single disease entity at the molecular level. Staudt
and colleagues have used hierarchical clustering to sub-
classify diffuse large B-cell lymphoma into three subtypes,
each characterized by specific groups of gene signatures
[14]. In addition to shedding light on the biology of dif-
fuse large B-cell lymphoma (DLBCL), this underscores the
fact that even using techniques with high statistical power,
the quality of unsupervised clustering and differential
expression analysis will depend largely on the underlying
assumptions about sample identity.
Genes with similar expression patterns are likely to have
related functions (‘guilt by association’ [15]), therefore,
finding genes with expression patterns similar to those of
known tumor markers is usually a convenient and effec-
tive method to employ in the search for new candidates.
This can be done by comparing the gene expression pat-
terns in a dataset to the pattern of the known marker in a
pair-wise fashion using an appropriate distance metric (e.g.
Pearson’s correlation coefficient) and choosing the high-
est-ranking hits. Even considering more comprehensive
unsupervised clustering approaches, this focused alterna-
tive way of ‘grabbing the low-hanging fruit’ still has merit,
especially when applied to data from new chips represent-
ing genes not present on prior chips.
Equally important to these various methodologies is the
accumulation of microarray data suitable for cancer target
discovery. Numerous gene expression studies in many can-
cer types are associated with entire datasets that can be
downloaded, many in public repositories specifically dedi-
cated to the dissemination of these valuable data. These in-
clude websites such as the Stanford Microarray Database
(http://genome-www5.stanford.edu/MicroArray/SMD), the
NCBI’s Gene Eexpression Omnibus repository (http://www.
ncbi.nlm.nih.gov/geo), the EBI’s ArrayExpress (http://www.
ebi.ac.uk/arrayexpress) and the MIT Cancer Genomics
Program (http://www.broad.mit.edu/cancer/). These are
valuable resources for target discovery. For example, the
Global Cancer Map data collection [16], which contains
data for 218 tumor samples of 14 common tumor types
along with 90 normal tissues, can be readily utilized for
finding genes highly expressed in lung cancer but low in
other normal tissues. There are still tremendous challenges
in integrating microarray data from different sources, as
microarray platform and data representation can be het-
erogeneous and inconsistent, but the standardization [17]
and integration [18] of microarray data should emerge in
the coming years to enable uniform representation of array
data and cross-platform data comparison for effective
target discovery purposes.
Finally, attractive targets are not defined solely by their
expression patterns. Other criteria, such as type of mol-
ecule (e.g. kinase), subcellular localization (e.g. cell surface)
and biological pathway (e.g. angiogenesis), are important
in the decision to follow-up on a potential new target. To
this end, resources such as the Gene Ontology Project
(http://www.geneontology.org) and the Kyoto Encyclopedia
of Genes and Genomes Pathways Project (http://www.
genome.ad.jp/kegg) attempt to place genes in the context
of biological function, location and pathway. The recent
W eb Server issue of Nucleic Acids Research [19] contains a
current overview of many tools that use these and other
resources.
Digital expression profiling using EST and SAGE
Gene expression profiling is not necessarily synonymous
with microarrays. ‘Digital expression’ based on either ex-
pressed sequence tags (ESTs) or serial analysis of gene ex-
pression (SAGE) is complementary to microarrays and can
be just as powerful. Both EST-derived expression and SAGE
are based on the principle that the frequency of sequence
tags sampled from a pool of cDNAs is directly proportional
to the expression level of the corresponding gene (see
Figure 2). Digital expression offers three major advantages
over microarrays. First, the simple digital data format in se-
quence clone counts and frequencies enables direct and
platform-independent data comparison among different
data collections from multiple tissues. Second, because
there is no need for designing any DNA chips, no prior
knowledge of gene sequence is required and therefore
many novel genes not covered by microarrays are repre-
sented. Finally, because the expression level is represented
by mRNA abundance relative to all transcripts and is inde-
pendent of probe selection and hybridization biases, digi-
tal expression can be a more quantitative measurement of
gene expression than microarrays [20].
Although ESTs were initially collected primarily for novel
gene identification [21], their value in gene expression
analysis became obvious with the accumulation of EST
data and development of appropriate computational tools.
There are currently over 5 million human EST sequences
available in the public database [22], derived from many
laboratories and tissue types over the last ten years.
Notably, a significant fraction of these ESTs are derived
from cancer tissues as a result of the large-scale efforts of
the Cancer Genome Anatomy Project (CGAP; http://www.
ncbi.nlm.nih.gov/ncicgap) at the National Cancer Institute
797
DDT Vol. 9, No. 18 September 2004 reviewsresearch focus

(NCI) to generate EST libraries from tumor samples [23].
Excluding normalized or subtracted cDNA libraries that are
not suitable for expression studies, ESTs present an attrac-
tive source for not only differential expression analysis be-
tween normal and cancer tissues but also for multi-tissue
expression profiling needed for toxicity prediction.
One common application of EST data is the pair-wise
comparison of libraries (or pools of libraries) of different
origins, for example, tumor versus normal libraries. Such
comparisons are facilitated by tools like Digital Differential
Display (DDD) [24] at the NCBI, cDNA Digital Gene
Expression Displayer (DGEN), and xProfiler at CGAP. The
reliability of this kind of expression result is largely depen-
dent on the size of the component libraries and can be
measured by Fisher’s exact test [24], the Z-test for two sample
proportions, and other statistical measurements [25].
Although pair-wise expression comparison is useful in
finding genes that are expressed at a higher level in tumor
than the corresponding normal tissues, cancer target discov-
ery requires comprehensive profiling of gene expression in
798
Figure 2. A schematic representation of several types of high-throughput data useful for cancer bioinformatics studies. (a) For digital
expression, collections of random clones harboring EST or SAGE tags are picked and DNA sequences determined. The expression level of a
particular gene is measured by the relative frequency of sequence tags for the gene. (b) Microarrays measure transcript abundance by
hybridizing fluorescently labeled RNA to DNA probes on a chip. Gene expression can be measured either on an absolute intensity scale or
as a ratio of intensities between a test (cancer) and a reference (normal) sample. Both digital and microarray expression data can be used
for large-scale differential gene expression analysis. (c) Array comparative genomic hybridization (CGH) is similar to expression microarrays,
except that instead of RNA, labeled genomic DNA is hybridized to DNA chips. In this case, the relative hybridization intensity of a cancer
DNA sample to a normal DNA sample (the reference) represents the relative abundance of genomic DNA in the cancer sample, which is a
measure of copy number change.
Drug Discovery Today
Tissues
cDNA
Cancer
Data
RNA
(c) Array CGH
Clone abundance →
Expression level
Absolute or relative
hybridization intensity →
Expression level
Relative hybridization intensity
→ DNA copy number changes
(a) Digital expression (b) Expression by
microarray
Reporting
system
Analyses
Test DNA Reference DNA
Hybridization to chip
Hybridization to chip
Random EST/SAGE
clone sequencing
Tag/library data Microarray data
Array CGH data
Normal
log2ratio
−3
−2
−1
0
1
2

a wide spectrum of normal and tumor tissues. To facilitate
this process, we have recently built a computational tool
named GEPIS (gene expression profiling in silico), which
profiles expression levels in multiple normal and tumor tis-
sues based on pools of EST libraries [20]. By iterative exam-
ination of such profiles for all available genes, we used
GEPIS to find desirable expression patterns – high expres-
sion in one type of tumor tissue but low elsewhere, includ-
ing the non-related normal tissue types. The expression
patterns of many of the targets identified by GEPIS have
been subsequently validated by other experimental meth-
ods, including quantitative RT–PCR and tissue microarrays.
Alternatively, as with microarray data, one could use guilt-
by-association methods to identify additional targets whose
expression patterns mimic those of known cancer-associated
genes [26].
Like EST-based expression analysis, SAGE relies on the
frequency of occurrence of sequence tags in a library, but
in this case the tags are 14-bp or 21-bp short sequences that
identify the cDNAs. The SAGE method enzymatically con-
catenates multiple such short sequence tags into each
clone, enabling many data points to be obtained from a
single sequencing read [27]. Subsequent analysis is similar
to EST-based methods. CGAP has sponsored the generation
of SAGE libraries in numerous normal and cancerous tis-
sues and makes the data available at the NCBI and CGAP
websites. Coupled with computational tools, such as
SAGEmap [28] from NCBI and SAGE Genie [29] from
CGAP, SAGE libraries have become an excellent source for
cancer-related expression analysis. SAGE data mining fol-
lowed by PCR screening was able to identify candidate
glioblastoma markers and antigens [30], and there are
799
Table 1. Representative online resources for cancer target finding and analysis
Name Description URL
Cancer Genes Collection of cancer genes based
on mutation data
http://www.sanger.ac.uk/genetics/CGP/Census/
Stanford Microarray Database
(SMD)
Repository of microarray data with
analysis tools
http://genome-www5.stanford.edu/
Whitehead Institute Center for
Genome Research, Cancer
Genomics Program
Repository of microarray data from
cancer genomics publications
http://www.broad.mit.edu/cgi-bin/cancer/
datasets.cgi
Gene Expression Omnibus
(GEO)
Repository of microarray data from
a variety of sources, including CGH
http://www.ncbi.nlm.nih.gov/geo/
Gene Ontology Project (GO) Controlled vocabulary describing
molecular function, biological
process, and cellular component
http://www.geneontology.org/
Kyoto Encyclopedia of Genes
and Genomes Pathways
(KEGG Pathways)
Molecular interaction networks,
including metabolic and regulatory
pathways, and molecular complexes
http://www.genome.ad.jp/kegg/metabolism.html
SAGEmap SAGE data repository and analysis
tools
http://www.ncbi.nlm.nih.gov/sage
SAGE Genie Analytical method of tag-to-gene
mapping with intuitive display
http://cgap.nci.nih.gov/SAGE
DGEN Digital gene expression display tool
that quantitatively compares two
pools of EST libraries
http://cgap.nci.nih.gov/Tissues/GXS
cDNA xProfiler Compares gene expression of two
pools of EST libraries
http://cgap.nci.nih.gov/Tissues/xProfiler
GEPIS EST-based gene expression profiling
data and analysis
http://www.cgl.ucsf.edu/Research/gene.ntech/gepis/
Progenetix Repository of cytogenetic
abnormalities in human cancer
http://www.progenetix.com/
SKY/M-FISH and CGH
Database
Repository of cytogenetic
abnormalities in human cancer
http://www.ncbi.nlm.nih.gov/sky/
SNP500Cancer Validated SNPs in cancer genes http://snp500cancer.nci.nih.gov/home.cfm

many other examples of the application of SAGE in cancer
target discovery [31]. Given the common principle of EST-
and SAGE-based expression and their digital data format,
computational tools can be developed that combine se-
quence tags between platforms to produce comprehensive
profiles that depict expression levels in thousands of
diverse tissues. The continual accumulation of ESTs and
SAGE tags, which currently stands at over 10 million, will
only rise in value as a resource for cancer target discovery.
Cancer association from recurrent DNA amplification
Part of the multi-step process of tumor formation is a pe-
riod of genomic instability usually resulting in regions of
genomic copy number increase. Oncogenes such as c-myc
have long been known to be associated with regions of
high copy number, and mapping such amplicons in tumor
cells has become a common method for searching for new
oncogenes or determining which known oncogenes might
be contributing to a particular cancer type. The methods
for detecting recurrent DNA amplifications have tradition-
ally been cytogenetic [comparative genomic hybridization
(CGH), M-FISH, SKY], and have resulted in the characteriz-
ation of many cancer-associated amplicons, albeit with res-
olution limited to what is detectable in a microscope (e.g.
the order of 2–10 Mb). Databases such as Progenetix.net
[32] and the NCI and NCBI’s SKY/M-FISH and CGH
Database have begun to serve as public repositories for
regions of chromosomal amplifications and deletions.
Recently, microarray technology has been adapted for
CGH (array-CGH, or matrix-CGH) to increase the resolu-
tion over the microscope-based approaches and to enable
much finer mapping of cancer amplicons [33]. Array-CGH
is similar in principle to array gene expression (see
Figure 2). The probes on the chip are mapped sequences,
usually BAC clones, although cDNAs and oligonucleotides
have also been used. The hybridizing test samples are
genomic DNA from tumor samples. Genomic DNA from a
normal sample serves as a universal reference, with the
test:reference ratio representing the relative copy number
of a given probe. With the emergence of array-CGH as a
high-throughput technology, and particularly as the chips
become denser and resolution improves [34], bioinformat-
ics approaches have enabled the identification of ampli-
cons that occur in many tumors. The area of minimal over-
lap of such recurrent amplicons greatly facilitates
identification of the oncogene candidates in those tumors
[35].
Like many genomic technologies, array-CGH is particu-
larly powerful when combined or integrated with comple-
mentary data from independent genomic approaches.
Gene expression tends to be higher for genes located
within amplicons [36,37], and array-CGH is often used in
conjunction with expression microarrays to identify likely
‘driver’ genes within amplicons [38]. Chip design is an im-
portant consideration for these studies because there is also
a subtle correlation between expression levels and probe
localization on chips [39]. In another approach, Zardo et
al. [40] have searched for potential tumor suppressor genes
in brain tumors by using restriction landmark genomic
scanning (RLGS) to identify areas of high methylation,
which were then compared to regions of genomic copy
number loss obtained through array-CGH analysis. Protein
expression on tissue microarrays [41], SKY [42] and con-
ventional CGH have all been used to supplement array-CGH
data and narrow down the field of potential oncogenes.
Gene expression data alone can also be used indepen-
dently to discover regions of genome amplification in silico.
In an EST-based whole-genome transcriptome analysis, the
Z-test was used to measure the extent of up-regulation in
cancer for a single gene, and a sensing index was designed
to scan for genomic windows where a cluster of genes
show higher expression in cancer [43]. The resulting non-
random Regions of Increased Tumor Expression (RITEs)
were found in all tissues examined, and they appear to
correlate with experimentally determined amplicons.
Similarly, when microarray-based expression data were
analyzed in the context of genomic organization, malig-
nancy-associated regions of transcriptional activation
(MARTA) were detected as spatial clusters of highly expressed
genes in various cancers [44]. In the absence of array CGH
data, both RITE and MARTA data provide rapid and valu-
able cancer amplicon leads, which tend to harbor genes
that are causally implicated in cancer.
Cancer gene finding from variant analysis
As a genetic disease, cancer arises due to the accumulation
of mutations in crucial genes that influence cell prolifera-
tion, differentiation and death. Identification of mutated
genes that are causally implicated in oncogenesis is an im-
portant aspect of target discovery. Either loss-of-function
mutations in tumor suppressors like p53 [45,46] or gain-of-
function mutations in oncogenes like BRAF [47] can play
promoting roles in oncogenesis. Identification of mu-
tations that occur predominately in cancer should help ex-
pand the collection of tumor suppressors and oncogenes,
which are often targeted for cancer therapy. In addition,
knowledge of cancer-causing mutations will facilitate early
diagnosis and help identify groups of patients who re-
spond better to a particular cancer treatment. Recently,
studies of EGFR somatic point mutations in lung cancer
patients underscore the link between clinical responses to
mutations in target genes [48,49].
800

Two types of cancer-specific genetic changes are of
particular interest: cancer-associated point mutations and
splicing variants. Although many point mutations have no
functional consequence, if one is found to be statistically
associated with the cancer phenotype it might then be
considered to be causal in cancer development. This is true
even if the functional consequence of the variant/mu-
tation is not immediately obvious, although non-silent
mutations are clearly the most attractive candidates for fol-
low-up. The same rationale applies to splice variants that
are predominately and non-randomly observed in cancer
samples. Compared with somatic point mutations, single
nucleotide polymorphisms (SNPs) are more frequent, ge-
netically stable, and often do not cause any deleterious
effect. Some SNPs might cause higher susceptibility to cancer
and are therefore expected to be overrepresented in cancer
samples. Here, we do not distinguish somatic mutations
from SNPs and refer to them collectively as cancer SNPs.
By linking tissue source, gene and sequence information
together, ESTs provide opportunities for in silico detection
of SNPs and splicing variants. Sequence alignments be-
tween ESTs and reference genes are often used for the ini-
tial detection of SNPs [50], and nearly 3 million SNPs have
been deposited into the NCBI’s dbSNP database [51]. With
EST libraries assigned with either normal or tumor sources,
statistical analyses such as Fisher’s exact test can be used to
determine whether a specific SNP is over-represented in
cancer [52]. Again, the same principles apply to the study
of cancer-specific splice variants. Although exon junction
microarrays are starting to show great utility [53], most
large-scale RNA splicing analyses rely on intron/exon struc-
tures revealed by EST sequences [54]. With EST source
information, every available human mRNA can be com-
pared with its corresponding normal and cancer ESTs using
sequence alignment software, and tumor-associated splice
forms were identified using the Z-test [55]. Alternatively,
with normal and cancer ESTs aligned to genomic sequences,
log odd ratio (LOD) can be calculated to detect splicing
forms overrepresented in cancer tissues [56]. It is worth
noting that although the in silico detection of gene vari-
ants holds great promise, it is subject to the same limita-
tions of all bioinformatics approaches, which is that the
results need experimental validation to avoid false leads
derived from noisy data. Nevertheless, the high-through-
put nature of in silico screening provides a valuable initial
step in finding mutated genes that are causally involved in
cancer.
Conclusion
As a genetic disease, cancer leaves a trail of genetic markers
accompanying tumorigenesis and cancer progression.
Somatic mutations, genomic instability and altered gene
expression patterns all provide possible ways to distinguish
cancer from normal cells, and such distinctions can help
us develop therapies that specifically target cancer cells. As
an enabling technology, bioinformatics has evolved in
many ways that enable us not only to identify players in
cancer pathways but also to comprehend genetic changes
in cancer.
Cancer target discovery requires integrated and high-
throughput analysis of genes, gene variants, expression
and DNA copy number changes. A recurrent theme for
bioinformatics data mining is that data from many differ-
ent sources continues to accumulate in public repositories.
As the amount and variety of the data continues to in-
crease, bioinformatics methods can continue to be devel-
oped, refined and applied to exploit the larger datasets.
Bioinformatics is not limited to a specific technology or
type of information, and in fact much of its utility lies in
integrating disparate data into a web of evidence used to
weigh the quality of potential targets. The application of
bioinformatics methods in cancer target discovery is start-
ing to generate many exciting target leads for further
experimental validation. Although bioinformatics-driven
target discovery is still in its infancy, it has already become
an indispensable piece of technology for cancer therapy
development in this post-genomic era.
Acknowledgements
The authors would like to thank Colin Watanabe, Thomas
Wu and Paul Polakis for critical review of the manuscript
and Allison Bruce for assistance in preparing the figures.
References
1 Green, M.C. et al. (2000) Monoclonal antibody therapy for solid
tumors. Cancer Treat. Rev. 26, 269–286
2 Futreal, P.A. et al. (2004) A census of human cancer genes. Nat. Rev.
Cancer 4, 177–183
3 Altschul, S.F. et al. (1990) Basic local alignment search tool. J. Mol. Biol.
215, 403–410
4 Eddy, S.R. (1998) Profile hidden Markov models. Bioinformatics 14, 755–763
5 Clark, H.F. et al. (2003) The secreted protein discovery initiative (SPDI),
a large-scale effort to identify novel human secreted and
transmembrane proteins: a bioinformatics assessment. Genome Res. 13,
2265–2270
6 Nielsen, H. and Krogh, A. (1998) Prediction of signal peptides and
signal anchors by a hidden Markov model. Proc. Int. Conf. Intel. Syst.
Mol. Biol. 6, 122–130
7 Krogh, A. et al. (2001) Predicting transmembrane protein topology with
a hidden Markov model: application to complete genomes. J. Mol. Biol.
305, 567–580
8 LeCouter, J. et al. (2001) Identification of an angiogenic mitogen
selective for endocrine gland endothelium. Nature 412, 877–884
9 Leung, Y.F. and Cavalieri, D. (2003) Fundamentals of cDNA microarray
data analysis. Trends Genet. 19, 649–659
10 Golub, T.R. et al. (1999) Molecular classification of cancer: class
discovery and class prediction by gene expression monitoring. Science
286, 531–537
801

11 Wu, T.D. (2002) Large-scale analysis of gene expression profiles. Brief.
Bioinform. 3, 7–17
12 Eisen, M.B. et al. (1998) Cluster analysis and display of genome-wide
expression patterns. Proc. Natl. Acad. Sci. U. S. A. 95, 14863–14868
13 Tusher, V.G. et al. (2001) Significance analysis of microarrays applied to
the ionizing radiation response. Proc. Natl. Acad. Sci. U. S. A. 98,
5116–5121
14 Alizadeh, A.A. et al. (2000) Distinct types of diffuse large B-cell
lymphoma identified by gene expression profiling. Nature 403, 503–511
15 Quackenbush, J. (2003) Genomics. Microarrays – guilt by association.
Science 302, 240–241
16 Ramaswamy, S. et al. (2001) Multiclass cancer diagnosis using tumor
gene expression signatures. Proc. Natl. Acad. Sci. U. S. A. 98,
15149–15154
17 Brazma, A. et al. (2001) Minimum information about a microarray
experiment (MIAME)-toward standards for microarray data. Nat. Genet.
29, 365–371
18 Detours, V. et al. (2003) Integration and cross-validation of high-
throughput gene expression data: comparing heterogeneous data sets.
FEBS Lett. 546, 98–102
19 Editors (2004) Editorial. Nucleic Acids Res. 32, W1
20 Zhang, Y. et al. GEPIS - quantitative gene expression profiling in normal
and cancer tissues. Bioinformatics (in press)
21 Adams, M.D. et al. (1991) Complementary DNA sequencing: expressed
sequence tags and human genome project. Science 252, 1651–1656
22 Boguski, M.S. et al. (1993) dbEST–database for ‘expressed sequence
tags’. Nat. Genet. 4, 332–333
23 Brentani, H. et al. (2003) The generation and utilization of a cancer-
oriented representation of the human transcriptome by using expressed
sequence tags. Proc. Natl. Acad. Sci. U. S. A. 100, 13418–13423
24 Scheurle, D. et al. (2000) Cancer gene discovery using digital
differential display. Cancer Res. 60, 4037–4043
25 Stekel, D.J. et al. (2000) The comparison of gene expression from
multiple cDNA libraries. Genome Res. 10, 2055–2061
26 Walker, M.G. et al. (1999) Prediction of gene function by genome-scale
expression analysis: prostate cancer-associated genes. Genome Res. 9,
1198–1203
27 Velculescu, V.E. et al. (1995) Serial analysis of gene expression. Science
270, 484–487
28 Lash, A.E. et al. (2000) SAGEmap: a public gene expression resource.
Genome Res. 10, 1051–1060
29 Boon, K. et al. (2002) An anatomy of normal and malignant gene
expression. Proc. Natl. Acad. Sci. U. S. A. 99, 11287–11292
30 Loging, W.T. et al. (2000) Identifying potential tumor markers and
antigens by database mining and rapid expression screening. Genome
Res. 10, 1393–1402
31 Porter, D. and Polyak, K. (2003) Cancer target discovery using SAGE.
Expert Opin. Ther. Targets 7, 759–769
32 Baudis, M. and Cleary, M.L. (2001) Progenetix.net: an online repository
for molecular cytogenetic aberration data. Bioinformatics 17, 1228–1229
33 Snijders, A.M. et al. (2001) Assembly of microarrays for genome-wide
measurement of DNA copy number. Nat. Genet. 29, 263–264
34 Ishkanian, A.S. et al. (2004) A tiling resolution DNA microarray with
complete coverage of the human genome. Nat. Genet. 36, 299–303
35 Albertson, D.G. et al. (2000) Quantitative mapping of amplicon
structure by array CGH identifies CYP24 as a candidate oncogene.
Nat. Genet. 25, 144–146
36 Pollack, J.R. et al. (2002) Microarray analysis reveals a major direct role
of DNA copy number alteration in the transcriptional program of
human breast tumors. Proc. Natl. Acad. Sci. U. S. A. 99, 12963–12968
37 Virtaneva, K. et al. (2001) Expression profiling reveals fundamental
biological differences in acute myeloid leukemia with isolated trisomy
8 and normal cytogenetics. Proc. Natl. Acad. Sci. U. S. A. 98, 1124–1129
38 Martinez-Climent, J.A. et al. (2003) Transformation of follicular
lymphoma to diffuse large cell lymphoma is associated with a
heterogeneous set of DNA copy number and gene expression
alterations. Blood 101, 3109–3117
39 Kluger, Y. et al. (2003) Relationship between gene co-expression and
probe localization on microarray slides. BMC Genomics 4, 49
40 Zardo, G. et al. (2002) Integrated genomic and epigenomic analyses
pinpoint biallelic gene inactivation in tumors. Nat. Genet. 32, 453–458
41 Zaharieva, B.M. et al. (2003) High-throughput tissue microarray
analysis of 11q13 gene amplification (CCND1, FGF3, FGF4, EMS1) in
urinary bladder cancer. J. Pathol. 201, 603–608
42 Harding, M.A. et al. (2002) Functional genomic comparison of lineage-
related human bladder cancer cell lines with differing tumorigenic and
metastatic potentials by spectral karyotyping, comparative genomic
hybridization, and a novel method of positional expression profiling.
Cancer Res. 62, 6981–6989
43 Zhou, Y. et al. (2003) Genome-wide identification of chromosomal
regions of increased tumor expression by transcriptome analysis.
Cancer Res. 63, 5781–5784
44 Glinsky, G.V. et al. (2003) Common malignancy-associated regions of
transcriptional activation (MARTA) in human prostate, breast, ovarian,
and colon cancers are targets for DNA amplification. Cancer Lett. 201,
67–77
45 Levine, A.J. et al. (1991) The p53 tumour suppressor gene. Nature 351,
453–456
46 Hollstein, M. et al. (1991) p53 mutations in human cancers. Science 253,
49–53
47 Davies, H. et al. (2002) Mutations of the BRAF gene in human cancer.
Nature 417, 949–954
48 Lynch, T.J. et al. (2004) Activating mutations in the epidermal growth
factor receptor underlying responsiveness of non-small-cell lung cancer
to gefitinib. N. Engl. J. Med. 350, 2129–2139
49 Paez, J.G. et al. (2004) EGFR mutations in lung cancer: correlation with
clinical response to gefitinib therapy. Science 304, 1497–1500
50 Brett, D. et al. (2000) EST analysis online: WWW tools for detection of
SNPs and alternative splice forms. Trends Genet. 16, 416–418
51 Sherry, S.T. et al. (2001) dbSNP: the NCBI database of genetic variation.
Nucleic Acids Res. 29, 308–311
52 Qiu, P. et al. (2004) Genome wide in silico SNP-tumor association
analysis. BMC Cancer 4, 4
53 Johnson, J.M. et al. (2003) Genome-wide survey of human alternative
pre-mRNA splicing with exon junction microarrays. Science 302,
2141–2144
54 Modrek, B. et al. (2001) Genome-wide detection of alternative splicing
in expressed sequences of human genes. Nucleic Acids Res. 29, 2850–2859
55 Wang, Z. et al. (2003) Computational analysis and experimental
validation of tumor-associated alternative RNA splicing in human
cancer. Cancer Res. 63, 655–657
56 Xu, Q. and Lee, C. (2003) Discovery of novel splice forms and
functional analysis of cancer-specific alternative splicing in human
expressed sequences. Nucleic Acids Res. 31, 5635–5643
802
Conference reports
Conference participants who wish to cover a particular meeting should contact:
Dr Jayne Carey, Drug Discovery Group, Elsevier, 84 Theobald’s Road, London, UK WC1X 8RR
e-mail: DDT@drugdiscoverytoday.com

Biotech2012spring 1-overview 0

More Related Content

What's hot (19)

Viewers also liked (14)

Similar to Biotech2012spring 1-overview 0 (20)

More from BioinformaticsInstitute (20)

Recently uploaded (20)

Biotech2012spring 1-overview 0