An Introduction To Genomics And SAS Scientific Discovery Solutions ...
An Introduction to Genomics and SAS® Scientific Discovery Solutions
Russ Wolfinger, SAS Institute Inc., Cary, NC
Kristen Quinlan, SAS Institute Inc., Cary, NC
Susan Flood, SAS Institute Inc., Cary, NC
The central dogma has been well-known for decades, so why
ABSTRACT
such a stir about it now? The answer is at the bottom of Figure 2.
Breakthroughs in microtechnological instrumentation, such as
The commoditization of genomics instrumentation has created a
DNA sequencers, microarrays, and mass spectrometers, have
strong need for management and analysis of the resultant high
revolutionized laboratory practice over the past few years,
volumes of complex data. This paper provides a brief description
transforming molecular biology from a data-poor to a data-rich
of the basic science underlying these data sources and an
enterprise. High-throughput sequencers create maps of the
overview of the initial steps SAS has taken to address them in
entire DNA blueprint of common organisms, including the 3 billion
our new SAS Scientific Discovery Solutions product line. Three
base pairs comprising the human genome. Microarrays provide a
core SAS foundational elements--data warehousing, analytical
snapshot of the number of gene transcripts in a given biological
servers, and JM--apply across the three principal foci of the
sample, enabling gene expression profiling across thousands of
central dogma of molecular biology: genetics, transcriptomics,
genes simultaneously. Mass spectrometers quantitate thousands
and proteomics. The integration of these elements in a unified
of protein levels from a sample using advanced laser and time-of-
application provides you with unprecedented power to intelligently
flight techniques. The data sets from these instruments, typically
extract knowledge from your genomics data and make
several megabytes from a single experiment, demand the use of
breakthrough discoveries. We highlight the technical aspects of
advanced computer science and analytical techniques for
this system using a microarray data example and then discuss
intelligent interpretation. This exciting conflux makes SAS an
several new development directions.
excellent software platform from which to conduct genomics
discovery.
INTRODUCTION
The central dogma of molecular biology (Figure 1) provides the
crucial scientific underpinnings of the content of this paper.
Deoxyribonucleic acid (DNA), in the form of the classical double-
helix built with the four amino acids adenine, cytosine, guanine,
and thymine, exists and self-replicates in the cell nucleus. Genes,
which are small segments of DNA, code for the transcription of
thousands of different forms of ribonucleic acid (RNA). RNA
molecules move across the nuclear membrane into the cell
cytoplasm and serve as translation templates for tens of
thousands of proteins. Gibson and Muse (2002) provide a more
in-depth overview of the intricacies involved in this fascinating
process.
Figure 2: The central dogma and its attending technologies drive
SAS Genomics.
[Note on terminology: We use “genomics” as a catch-all word to
describe the aforementioned interplay between molecular
biochemical data and analytical software. Another popular term
is “bioinformatics,” which is often used in the same context.]
What specific SAS technologies are applicable to genomics?
Considering the breadth and depth of difficulties faced in this
field, nearly all of them have bearing. It is difficult to know where
to start. To provide appropriate focus, scope, and relevance for
this paper, we concentrate on an example from the second stage
of the central dogma, transcriptomics, and the key constituents of
the initial release of the SAS Scientific Discovery Solutions (SDS)
bundle. Figure 3 shows the foundation of this bundle as SAS
Research Data Management (RDM), a Java-based graphical user
interface to SAS/Warehouse Administrator that enables you to
extract, transform, load, and manage genomics research data in
a flexible and open fashion. The first vertical application of RDM
Figure 1: The Central Dogma of Molecular Biology
is the SAS Microarray Solution (MAS), designed specifically for
data on gene transcription. It includes input engines and
analytical processes for various kinds of microarray data, and is
also open and customizable. Examples and details about RDM
1
and MAS follow in the next two sections. The final section covers
Hsieh et al. (2002) for details about the other example, from the
several new applications and future development directions for
Affymetrix Latin Square experiment, and to Chu et al. (2002a,
SAS SDS.
2002b) for issues surrounding Affymetrix data analysis in general.
SAS RESEARCH DATA MANAGEMENT
Data
The vast amount of data generated by genomic experiments
engines
Future solutions
presents challenges in terms of managing and sharing data within
SAS
Data
models
and among organizations. Researchers need a direct, non-
Scientific
Visual’n
programming approach for acquiring data for analysis.
Discovery
Furthermore, numerical data are not the only information of
Analysis
Genetic
interest. You may also want to view images, documents, and
Solutions
SAS
Proteomics
HTS
Markers
Microarray
analysis results associated with the data.
Solution
(2002)
Cohesive solutions
SAS RDM is the data management core of the SAS Scientific
for discovery
SAS Research Data Management
Discovery Solutions. It is a Java client-server application that
research
utilizes SAS warehousing technologies to provide a centralized
Client
SAS R
Java A
esearch pp
D licatio
ata Mnanagement JMP
(2002)
Server
SAS Technologies (WA)
repository for your organization's discovery research data, and
other ancillary information.
DATA WAREHOUSING
The concept behind data warehousing involves extracting data
Copyright © 2002 , SAS Institute Inc. All rights reserved.
Figure 3: RDM as a Foundation for SAS SDS
and reorganizing that data in a form better suited to analysis and
reporting. The key to data warehousing is metadata, or data
about data. Examples of metadata include a data file name,
EXAMPLE: TRANSCRIPTION IN DROSOPHILA
location, storage format, and structure. A data warehouse
Jin et al. (2001) describe results of a microarray experiment for
captures metadata throughout the warehousing process: where
assessing transcriptional changes in fruit flies. Twenty-four two-
the data came from, what transformations were performed, and
color arrays, each probing approximately 3,000 genes, reveal
information about the context of the data. Managing the metadata
gene expression differences between two lines (Samarkand and
and controlling the transformation processes are the key benefits
Oregon), two genders (Female and Male), and two ages (1 week
of data warehousing.
and 6 week). Figure 4 shows the basic steps for generating data
POOLED METADATA REPOSITORY
from a single, generic, two-color array. Note that with this kind of
array, two signals are generated for each gene on each array
SAS RDM provides a platform for centralizing access and
(one from the Cy3 fluorescent dye and one from the Cy5
managing this discovery research data. At its core is the Pooled
fluorescent dye), so a split-plot arrangement is possible for the
Metadata Repository (PMR), which provides a means of
three experimental factors.
consolidating metadata from any number of data warehouses to
create a single searchable repository. The PMR does not move
the physical data, but rather gathers distributed metadata into
one location
With the PMR, RDM extends the data warehousing concept to its
scientific users by providing additional capabilities that enable
compliance with FDA regulations, promote collaboration between
departments/projects, organizations and/or locations, and allow
the management of multiple object types used in discovery data.
The RDM workspace surfaces the PMR to researchers in the
form of searchable metadata. Figure 5 illustrates such a search
on the data from the Drosophila example. You can search and
download all information associated with a project or an
experiment, including data, documents, and images.
In addition to the search and download capabilities, you can
register new data to the PMR to share with colleagues via the
Upload component. Upload functionality provides a way for you to
Figure 4: Data Generation Process for a Two-color Microarray
add individual objects to the PMR one at a time. Figure 6 depicts
upload of a PDF document related to the Drosophila experiment.
Since primary interest is in the aging effect, Jin et al. consider
Once you upload an object, it becomes available to anyone who
age (and dye, by necessity) to be the subplot factors, and line
has access to the PMR. You can add extended attributes to any
and gender to be the whole-plot factors. This experimental
object, which then become searchable metadata.
design is much more efficient than the popular reference sample
design, in which a noninformative reference sample is always
hybridized in either the Cy3 or Cy5 channel (Kerr and Churchill,
SECURITY MODEL
2001). In addition, this design enables simultaneous assessment
Among the regulatory compliance features is a security model
of the main effects and interactions of the age, line, and gender
that controls access to the system by requiring a user name and
effects on each gene, all adjusted for potential dye effects.
password. RDM controls access to warehouse objects as well as
access to certain areas of the system at the user level or at a
We use data for 100 genes from this experiment as a running
user-group level. Permissions range from no access to read-only
example below. This small subset is one of two test examples
access to full edit access. The security model also includes an
that ship with the SAS Microarray Solution software. Refer to
audit trail of actions performed in the system (Figure 7) and
versioning of all data loaded in the warehouse. The audit trail and
version control simplify the ability to trace back the source of data
2
and/or modifications made along the way.
Figure 5: Perform a search on the Pooled Metadata Repository.
Figure 7: SAS Research Data Management Audit Trail
SAS MICROARRAY SOLUTION
SAS MAS is the first vertical deployment of the RDM platform,
and integrates seamlessly with it to produce a whole product
solution for managing, analyzing, and visualizing microarray data.
The additional functionality of MAS includes input engines and
analytical processes.
INPUT ENGINES
Input engines are software routines that pull data from
instrumentation systems output and load the data into a SAS
MAS warehouse. Input engines are specific to input data
structures, and you can customize them to suit the vendors you
select to perform your microarray experiments. Input engines
enable you to directly import an entire collection of raw numerical
array files with a few mouse clicks. Figure 8 shows the input
engine for the Drosophila aging data, which were produced in the
Gibson Lab (http://statgen.ncsu.edu/ggibson/) using the popular
public domain image analysis software ScanAlyze
(http://rana.lbl.gov/EisenSoftware.htm).
Figure 6: Upload a file to the Pooled Metadata Repository.
DATA MODELS
Built using add-ins to SAS/Warehouse Administrator, RDM
enables you to implement virtually an unlimited number of data
schemas. For microarray data such as those from our
Drosophila example, the recently OMG-sanctioned MAGE-ML
standard (http://www.mged.org) provides a useful foundation. You
can use the SAS XML engine (Friebel, 2003) to import data from
MAGE-ML. Systems engineers from SAS can assist you with
this as a part of their warehousing pilot program.
Figure 8: ScanAlyze Input Engine for Drosophila Aging Data
In addition to the raw data files, you must provide an
experimental design file that shows how to map the experimental
factors to the arrays. Figure 9 shows a portion of the
experimental design table for the Drosophila experiment. Note
there are two rows for every array corresponding to the Cy3 and
3
Cy5 channels. The three experimental factors are Line, Sex, and
experiment, divides it into groups selected by the user, and
Age, and there are also variables indicating the name of the
then performs a multivariate correlation analysis on each
corresponding raw data file and the appropriate column within it.
group.
This kind of approach allows you to input designs of arbitrary
• MixedModelNormalization normalizes microarray data by
complexity.
fitting a mixed linear model across all of the arrays in an
experiment.
• MixedModelAnalysis provides a comprehensive look at
results from fitting mixed models on a gene-by-gene basis.
ARRAY GROUP CORRELATION
Let’s consider ArrayGroupCorrelation in more detail. This
process accepts as input a SAS dataset in “tall skinny” format;
that is, all of the raw array intensity measurements are stacked
into one variable. (The default input engines build this kind of
dataset for you.) It then enables you to group the observations
into sets that will be plotted against each other in multivariate
fashion.
Figure 10 displays the parameter input window for
ArrayGroupCorrelation. MAS creates this window dynamically
from specially configured SAS macro code. The SAS Microarray
Solution Analytical Process Programmer’s Guide, available only
to SAS MAS customers, describes the requisite details. You
must enter appropriate information in each field and then click
Submit.
Figure 9: Experimental Design Table for the Drosophila Example
A different input engine enables you to input annotation data for
the genes on a particular chip, and you can use the resulting SAS
dataset in any experiment for which it is appropriate.
ANALYTICAL PROCESSES
Analytical processes are special SAS macro programs that
perform data manipulations and statistical calculations on the
experimental data you have loaded into RDM. These processes
employ the power of the SAS System to generate analysis data
sets, listings, statistical results, and graphs. Analytical processes
are reusable and flexible. They can range in functionality from
very simple data displays to complex statistical modeling.
Scientists and statisticians alike can run analytical processes
against any experiment dataset by simply providing appropriate
values for input parameters. After you select an analytical
process to run, a parameter input window requests information
required for successful execution. The parameters values are
specific to the dataset at hand, but you do not need to edit the
code itself once you have written and loaded the analytical
process into SAS MAS.
In its initial release, SAS Microarray Solution provides four
analytical processes for use with your experiment data:
Figure 10: Input Parameters for the ArrayGroupCorrelation
Analytical Process
•
DataContents displays the contents of a SAS dataset in
Upon submission, ArrayGroupCorrelation sends its macro code
HTML format.
via SAS Integration Technologies to a preactivated SAS server,
• ArrayGroupCorrelation takes all of the array data from an
assigning each parameter value from the input window to its
4
corresponding macro variable. The code performs some error
checking and then calls Proc Transpose to put the data into the
appropriate multivariate form. It then uses PUT statements to
create a JMP Scripting Language (JSL) file. SAS MAS executes
this file, producing output as in Figure 11. In addition to the
standard Multivariate platform for each group, this particular script
also creates a Web Search Dialog in JMP that allows you to
automatically display the relevant UniGene
(http://www.ncbi.nlm.nih.gov/UniGene/ ) or FlyBase
(http://flybase.bio.indiana.edu/) Web pages for all genes that are
currently selected.
Figure 11: JMP results from ArrayGroupCorrelation
Figure 12: Parameters for MixedModelAnalysis Analytical
Process
MIXED MODEL ANALYSIS
MixedModelAnalysis, like ArrayGroupCorrelation, executes on the
MixedModelAnalysis is a more complicated analytical process. It
SAS server and then creates a JSL file. Figure 13 shows some
performs a high-level mixed model analysis of variance on pre-
of the results for the Drosophila example. The upper-left window
normalized array data. (MixedModelNormalization implements
is a volcano plot, which graphs log2 fold change on the x-axis
one basic kind of normalization that centers each channel to its
versus statistical significance, in terms of negative log p-value, on
geometric mean.) Figure 12 shows the top portion of the
the y-axis. MixedModelAnalysis creates a separate volcano plot
MixedModelAnalysis parameter input window.
for each ESTIMATE statement you specify. The lower-left plot is
a parallel coordinate plot of the lsmeans of all genes that pass a
The critical input parameter is a collection of Proc Mixed
Bonferroni cutoff in at least one of the volcano plots. The bottom
statements enclosed in the %str() macro to allow SAS
middle plot are the same lsmeans, but standardized to have
punctuation. This code illustrates a full three-way factorial model
mean zero and variance one. The unstandardized lsmeans are
adjusted by a main effect for dyes. The process fits this model to
related to the x-axes of the volcano plots, whereas the
each gene separately using a BY statement with the GeneVar
standardized profiles are more closely related to the y-axes
parameter value as the SAS variable. Array (equivalent to spot at
because the t-tests behind the negative log p-values are location-
the individual gene level) is considered to be a random effect,
scale invariant. The upper-center graph plots the first two
and accounts for the typically strong correlation observed
principal components of the standardized lsmeans of the
between two measurements from the same spot. (Note that this
significant genes.
model is on the normalized log2 intensities, not ratios.) The three-
way interaction forms least-squares means, and a series of
The final display in Figure 13 is a two-way Wald hierarchical
ESTIMATE statements test various one degree-of-freedom
clustering analysis of the standardized lsmeans of the significant
hypotheses of interest. Some references for this kind of
genes. Rows are genes and columns are lsmeans category.
approach are Deng et al. (2002), Gibson (2002), Jin et al. (2001),
Although not legible, the left portion of this plot contains the
and Wolfinger et al. (2001).
annotation information for each gene. The colors in all of the
displays are derived from this clustering analysis. The dynamic
linking and interactivity of JMP make it ideal for intense
exploration of these statistically curated results.
5
Furthermore, you can use Text Miner to process collections
of scientific abstracts you create from important gene lists.
• SAS/MP-Connect enables parallelization of suitable
methods across multiple CPUs. Refer to Doninger et al.
(2003) for a recent microarray example.
BIOINFORMATICS
While SAS’s core strengths remain in warehousing and advanced
analytics, we are developing new tools in association with RDM
and MAS for such areas as motif detection, pathway overlays,
and enhanced annotation from public databases.
PROTEOMICS
Proteomics is an area of active research and development.
Neville et al. (2002) describe some initial analyses of mass
spectrometry data using both a data mining approach and a more
informal method based on Welch t-tests. For the latter, Figure 14
depicts overlaid blood serum spectra from a group of normal and
diseased patients, along with a negative log10-pvalue spectrum
indicating significant “hot spots” in the spectra above it.
Significant group-wide differences could translate into critical
diagnostic biomarkers after suitable validation.
Figure 13: Results of Mixed Model Analysis in JMP
The other two prewritten analytical processes, DataContents and
MixedModelNormalization, do not generate output in JMP.
Instead, they create HTML output via the Output Delivery System
that you can view in a browser. Not only are these four
processes useful for array analysis, but they provide you with
templates for writing your own processes.
NEW DEVELOPMENT DIRECTIONS
TRANSCRIPTOMICS
The preceding examples represent a small fraction of the
capabilities SAS offers you for analyzing microarray and related
transcriptomic data. RDM and MAS provide a framework that
enables you to tap into the full power of the SAS System in order
to make the most of your array data. Consider the following
Figure 14: Log spectra (Y) of diseased (red) and normal (green)
capabilities, all of which you can utilize via MAS Analytical
blood sera, along with negative log10 p-values (gray) testing for
Processes:
significant differences between the groups at each time-of-flight
(tof) value.
• The SAS Data Step, SAS Macro Language, and JMP
Scripting Language are very flexible and extensive
GENETICS
environments for manipulating and processing array data.
This paper ends where it started, with a consideration of DNA.
• Proc Optex, Proc Factex, Proc Plan, the ADX Menu System,
SAS/Genetics, introduced last year, is a standalone product
and the DOE features in JMP help you to create optimal
containing procedures for the statistical processing of DNA
designs.
marker data (Czika et al., 2001). New enhancements for Release
• Over half of the 50+ procedures in SAS/Stat are applicable
9.1 include the following:
to array data. These include procedures for analysis of
• Two new procedures are available: HTSNP (for finding
variance, clustering, density estimation, discriminant
haplotype tagging single-nucleotide polymorphisms) and
analysis, multidimensional scaling, multiple comparisons,
INBREED (copied from SAS/STAT).
nonparametric statistics, partial least squares, power and
sample size, principal components, and smoothing.
• The GENOCOL and DELIMITER= options allow you to input
(Rodriguez 2003). New residual diagnostics will be available
marker data with single variables instead of two variables
in Release 9.1 of Proc Mixed, enabling better quality control
per maker.
of your array data. Procedures from SAS/QC, SAS/OR,
• The ALLELE procedure now has the ALLELEMIN=,
SAS/ETS, and SAS/IML (including IML Workshop) can also
GENOMIN=, and HAPLOMIN= options for specifying
be useful. Scores of examples from other scientific
minimum estimated frequencies.
disciplines are in the SAS Sample Library and at
• The CASECONTROL procedure has the new NULLSNPS=
www.sas.com/techsup/download/stat .
option for specifying SNPS for genomic control and the
• You can use RDM and MAS to preprocess and create
PERMS= option for computing permutation-based exact p-
analysis-ready data sets for Enterprise Miner. Its incredibly
value approximations.
powerful collection of data mining methodologies and
• The FAMILY procedure has a new “Family Summary” table
intuitive process-flow interface enable you to efficiently
valuable for checking for inconsistencies, a PERMS= option
generate a wide range of cross-validated predictions.
like that in CASECONTROL, and multiallelic SDT and
6
combined SDT/TDT options.
Friebel (2003), XML? We do that! SUGI 28 Proceedings, SAS
• The HAPLOTYPE procedure has a new stepwise EM
Institute, Inc., Cary, NC.
algorithm.
Gibson, G. (2002), MMANMADA Tutorial,
• The PSMOOTH procedure has new TPM and TAU= options
http://statgen.ncsu.edu/ggibson/Pubs.htm .
for implementing the truncated product method.
Gibson, G. and Muse, S. (2001), A Primer of Genomic Science,
Details about these and other new features are available from
Sinhauer.
SAS technical support. We are also working on an integrated
genetic markers solution, similar in design to SAS Microarray,
Hsieh, W.-P., Chu, T.-M., and Wolfinger, R.D. (2002), Who are
that incorporates macros derived from SAS/Genetics in place of
those strangers in the Latin square? manuscript to be
the microarray best practices.
published in CAMDA proceedings, Duke University.
CONCLUSION
Jin, W., Riley, R., Wolfinger, R.D., White, K.P, Passador-Gurgel,
G. and Gibson G. (2001), Contributions of sex, genotype
Rich scientific data from microtechnological instrumentation and
and age to transcriptional variance in Drosophila
SAS software are a formidable combination. Regardless of
where along the central dogma your interests lie, SAS Scientific
melanogaster, Nature Genetics, 29:389-395.
Discovery Solutions equip you with the power to intelligently
Kerr and Churchill (2001), Experimental design for gene
manage and analyze your data with unparalleled proficiency.
expression microarrays, Biostatistics, 2:183-201.
http://www.jax.org/research/churchill/pubs/index.html
SAS RDM combines the power of data warehousing with an
Neville, P., Tan, P.-Y., Mann, G., and Wolfinger, R.D. (2002),
easy-to-use interface for search and retrieval. Its warehousing
Generalizable mass spec mining and mapping, manuscript
technologies can manage many different types of data,
to be published in proceedings of First Annual Proteomics
regardless of their formats or type. With SAS RDM, you can
Conference, Duke University.
register genetic, transcriptomic, protein and related data files to
the warehouse and view them using their native applications.
Rodriguez, R. (2003), SAS/STAT Version 9: Progressing into the
SAS RDM increases the productivity and effectiveness of your
Future, SUGI 28 Proceedings, SAS Institute, Inc., Cary, NC.
discovery organization by streamlining data management and
increasing data accessibility.
Wolfinger, R.D., Gibson, G., Wolfinger, E.D., Bennett, L.,
Hamadeh, H., Bushel, P., Afshari, C., and Paules, R.S.
SAS Microarray builds on the strong RDM foundation and
(2001), Assessing gene significance from cDNA microarray
enables you to tap into the power of the entire SAS System with
data via mixed models, Journal of Computational Biology, 8,
its rich data manipulation and analytical capabilities. It is
625-637.
designed for both statisticians and scientists to work in a
collaborative format. Statisticians can guide the scientist on
experimental design and write appropriate analytical processes.
ACKNOWLEDGMENTS
They can then register these processes into the solution where
The SAS Genomics R&D Team: Carey Carpenter, Tzu-Ming Chu,
scientists can easily access, use, and reuse them with no
Wendy Czika, Samuel Johnson, Geoffrey Mann, Chandrika
additional SAS programming. This enables both the scientist and
Nayak, Kristen Quinlan, Russ Wolfinger, Xiang Yu, and Jun
the statistician to spend more time doing high knowledge tasks
Zhang. RDM and MAS Testers: Shibing Deng and Kelly Graham.
that require their expertise.
SAS/Genetics Testers: Gerardo Hurtado and Jack Berry.
Of course software is no substitute for painstaking research,
It is infeasible to list here the scores of folks who have helped
careful thought, and effective personal collaboration with
make SAS SDS and this paper a reality, although we would like
colleagues. It can, however, significantly enrich these activities,
to thank the following individuals for their singular contributions:
and it has become indispensable as genomics information
Darrell Barton, Kurt Brumbaugh, Michael Campa, Bob Carpenter,
volume and density increase. We look forward to helping you
Julia Carpenter, Rob Carscadden, Joe Carter; Gary Churchill,
overcome the inherent challenges and achieve incredible new
Virginia Clark, Glen Dalrymple, Cheryl Doninger, Alan Eaton,
insights in scientific discovery.
Rick Evans; Michael Fitzgerald, Tony Friebel, Jill Fitzgibbons,
Greg Gibson, Russell, Gonsalves, Gerhard Held, Bob Hickey,
Ralph Hollinshead, James Holman, Hans Meyer, Joe Mudd,
Meltem Narter, Padraic Neville, Kay Obenshain, Ned Patz, Jon
Pennino, Bob Rodriguez, John Sall, Pei-Yi Tan, Jill Tao, Randy
REFERENCES
Tobias, Len van Zyl, Bruce Weir, Fred Wright, and Liz Wolfinger.
Chu, T.-M., Weir, B., and Wolfinger, R.D. (2002a), A systematic
statistical linear modeling approach to oligonucleotide array
CONTACT INFORMATION
data analysis, Mathematical Biosciences, 176, 35-51.
We value and encourage your comments and questions. Please
Chu, T.-M., Weir, B., and Wolfinger, R.D. (2002b), Comparison of
contact Susan Flood at:
Li-Wong and mixed model approaches to oligonucleotide
SAS Institute Inc.
array data analysis, in review.
SAS Campus Drive
Cary, North Carolina 27513
Czika, W., Yu, X., and Wolfinger, R.D. (2002), Genetic data
analysis using SAS/Genetics, SUGI 27 Proceedings, SAS
Phone: (919) 677-8000
Institute, Inc., Cary, NC.
Email: Susan.Flood@sas.com
Web: www.sas.com
Deng, S., Chu, T.-M., and Wolfinger, R.D. (2002), Transcriptome
variability in the normal mouse, manuscript to be published in
SAS and all other SAS Institute Inc. product or service names are
the CAMDA proceedings, Duke University.
registered trademarks or trademarks of SAS Institute Inc. in the
Doninger et al. (2003), Developing Client/Server Applications to
USA and other countries. ® indicates USA registration.
Maximize V9 Parallel Capabilities, SUGI 28 Proceedings,
SAS Institute, Inc., Cary, NC.
Other brand and product names are trademarks of their
respective companies.
7
