PNNL-13501 - Pacific Northwest National Laboratory

Bioinformatics for High-Throughput Proteomic and Genomic Analysis
Study Control Number: PN00015/1422
William Cannon, Heidi Sophia, Kristin Jarmin
Our goal is to develop the underlying science and technology for analyzing high-throughput genomic and proteomic data,
and for identifying proteins and elucidating the underlying cellular networks. Biology as a system has a logical and
mathematical foundation. Proteomics and DNA microarrays will provide the prerequisite data necessary for such a
systematic understanding of the cell.
Project Description
The long-term objective for this project is to provide the
tools for analyzing proteomic data from high-throughput
mass spectrometry experiments and for elucidating
cellular networks. Initial work focused on 1) developing
optimal clustering tools for examining either proteomic
data or DNA microarray data, 2) creating a library of
graph algorithms for use in the network analysis, and
3) for conceptual development of probability networks for
analyzing proteomic data.
Introduction
The goal of this project is to develop the underlying
science and technology for analyzing high-throughput
genomic and proteomic data, especially elucidating the
underlying cellular networks. We are taking a twopronged
approach. First, statistical and/or mathematical
modeling of the biological data in the form of
protein/gene expression levels will inferentially determine
networks of interactions. Second, previous biological
knowledge in the form of known regulatory motifs at the
DNA and protein levels will be incorporated to
complement the mathematical modeling. These tools will
be usable in isolation or in combination. It is anticipated
that the combinatorial use of these tools will provide the
synergy to bootstrap our way to biological networks much
more powerfully than the use of either set of tools in
isolation. The tools developed in the initial stage of the
project will be expanded in subsequent stages to
1) incorporate the ability to statistically infer cellular
networks from high-throughput data and 2) to sort the
high-throughput data by cellular regulatory function as
determined by, for example, cis-regulatory elements of
DNA and signal peptides.
Results and Accomplishments
New Clustering Techniques
The development of new clustering techniques for
proteomic data has focused on the use of NWGrid to
generate clusters. The distance metric is based on
Delauney triangulation and Voronoi polyhedra routines
within NWGrid. Initial results using a fixed number of
clusters have been encouraging. We are now proceeding
to develop methods by which the program can determine
the optimal number of clusters from the data itself.
Development of Software Library of Graph Algorithms
for Network Analysis
Work to date has been in optimizing existing algorithms
for use with expression data, and in making robust
versions of the algorithms.
Conceptual Development of Probability Networks for
Analyzing Proteomic Data and Strategic Planning
Much of the time during this period was spent developing
the concepts needed for long-term success. This includes
expanding the proposal with clear near- and long-term
goals. Also, a considerable amount of time is being spent
early in this project to develop a probability network
model that will serve as the foundation of our cellular
networks. Issues regarding connectivity of the network,
nature of the data, conditional independence of data are
being addressed.
Summary and Conclusion
In establishing a project foundation, we initiated new
clustering algorithms and a software library for analyzing
graph structures.
Computer Science and Information Technology 155