Jump-Starting Your Bioinformatics Career as an Undergraduate: One Student's Approach
Introduction
Bioinformatics is a multidisciplinary research field where computer science, applied mathematics, informatics, and statistics are used together to solve complex biological problems [11]. Perhaps the most compact definition would be to say that, "bioinformatics is the science of refining biological information into biological knowledge" in a computational framework [4].
In this article, we will see how an undergraduate student with computer science and engineering background may jump-start his or her training in the field and contribute to the advancement of science, as well as open the door to even bigger contributions in the future. To illustrate this, I will describe my own bioinformatics research experience over a one year period of time. During that time, I developed software to graphically represent gene regulatory networks for biomedical scientists.
Bioinformatics Research: Graphical Representation of Gene Regulatory Networks
A gene regulatory network (GRN) is a group of specific genes in which individual genes influence or regulate the activity of others. GRNs represent an apex in bioinformatics and computational biology, allowing for the potential discovery of triggering mechanisms and treatments for high-profile diseases.
In an effort to graphically represent GRNs, I developed a Java program entitled "ArrayViz"[10]. While specifics of the program will be discussed later in the article, in general, ArrayViz reads in gene expression microarray data, processes it statistically, and finally displays a graphical representation in the form of an undirected graph. The combination of skills required to complete this process draw from computer science and engineering, mathematics and statistical analysis, cellular biology, and especially the communication and interactions between those disciplines. In the end, the sampling of experience from each of these fields allowed me, as an undergraduate, to construct this tool.
Multidisciplinary Training Required for a Multidisciplinary Project
Through a mentor-guided undergraduate research program at Arizona State University[5], I was able to begin the multidisciplinary learning process in earnest. The following sub-sections detail my learning experiences in the fields of cellular biology, mathematics and statistics, and computer science, as well as their relation to ArrayViz, which is illustrated in Figure 1.
Figure 1: Multidisciplinary makeup of ArrayViz.
Cellular Biology
As a neophyte to cellular biology, I first had to learn
the basics of cellular function, DNA, DNA transcription,
RNA, genomics, the Human Genome Project, and finally how
they all fit together to produce the data set I had in my
hands.
I picked up a lot of jargon from various books and articles, but perhaps what helped the most was seeing it all put into practice at the Translational Genomics Research Institute (TGen) [9] in Phoenix, Arizona, where my aforementioned mentor works as a researcher.
The key to producing my data set is the DNA microarray. These microarrays are a large set of cloned DNA molecules spotted onto a solid matrix (such as a microscope slide) for use in probing a biological sample to determine gene expression, marker pattern, or nucleotide sequences of DNA/RNA. One DNA microarray chip can provide expression levels for anywhere from several hundred genes to hundreds of thousands of genes. After the microarray probes are analyzed and expression levels are determined, the final microarray data is made available in spreadsheet form.
Applied Mathematics and Statistics
The first step of the ArrayViz software involves reading
in the DNA microarray data and performing some
user-directed data transformation and normalization.
Transforming and normalizing a data set aids in
extrapolating the desired data while reducing statistical
noise. Transforming any data set logarithmically, for
example, will expand the scale of low numbers while
compressing the scale of higher ones. In terms of gene
expression data, this transformation will allow more
accurate visualization of different kinds of data, since
sets with a large range of intensities can be transformed
into more usable numbers.
ArrayViz next compares every pair of genes in the data set to determine their influence on one another, be it insignificant or significant. Depending on the number of genes in a data set, this step can be considerable in size, since the number of gene pairs follows the binomial coefficient, or "n choose k." ArrayViz currently allows two different statistical metrics for determining inter-gene influence. In many data sets, the expression readings contain discrete values over a large range with many decimal places. In other data sets, the expression values might be binary (0 or 1) or ternary (-1, 0, or 1) in form. These different data benefit from different metrics for statistical comparison, and ArrayViz currently offers the correlation coefficient and mutual information as metrics of statistical comparison between genes.
Computer Science
Based on the sheer number of gene-to-gene comparisons
needed for one graphical representation, it is easy to see
why computer science plays such an integral part in
bioinformatics in terms of efficiency alone. In the bigger
picture, the computational demands of ArrayViz are nearly
infinitesimal when compared with many of the jobs
researchers need to run on a regular basis. While ArrayViz
may pause for one or two seconds during computation, some
programs require weeks or even months to process, and
these programs run on some of the largest computers in the
world. By adhering to fundamental computer science
paradigms, ArrayViz was developed efficiently, with
special detail paid to execution efficiency.
The urgency to write good code is perhaps best expressed by the resources in which people are willing to invest in order to run large-scale bioinformatics software. Arizona State University and TGen recently partnered up to create one of the fifty fastest and largest supercomputers in the world. The 512-node cluster of IBM eServer xSeries 1350 servers powered by 1024 Intel Xeon processors is housed on the ASU campus. It runs Red Hat Linux and will eventually support hundreds of users and researchers on the system [7], including those from TGen, the Biodesign Institute[2], and the Department of Biomedical Informatics[3].
ArrayViz Functionality
With sufficient training in the areas of cellular biology, mathematics and statistics, and computer science, I was ready to begin work on ArrayViz in earnest. Many aspects of the software have been briefly mentioned to supplement the descriptions of the multidisciplinary aspects of the project. However, in order to offer a clearer understanding of how the software works, I present the following walk-through, which is comprised of six steps. These six steps, summarized in Figure 2, encapsulate the multidisciplinary communication and interaction required for bioinformatics research in the context of the ArrayViz software.
Figure 2: ArrayViz program flow.
Step 1: Microarray Data Input
Once a biologist has completed a microarray-based
experiment, he/she converts the data to an electronic
format, such as a spreadsheet. In certain free online
databases, such as Gene Expression Omnibus (GEO)[6], researchers can access microarray data from
countless other experiments. When the desired data has
been obtained, the researcher or biologist can hand it
over to ArrayViz.
Step 2: Mathematical Data
Transformation
We discussed how transforming and normalizing microarray
data can be helpful in extrapolating the desired
information while reducing noise. In the next step,
ArrayViz allows the user to transform the data set using
the log-base-two transform.
Step 3: Statistical Gene Comparison
After any desired transformations are complete, the next
task is to find any present GRNs. To do this, it is
necessary to compare every unique gene pair in the data
set using one of the statistical metrics previously
mentioned, namely, the correlation coefficient or mutual
information.
Step 4: Layout Generated by
GraphViz
After the interaction level of each gene pair has been
calculated, ArrayViz creates a specially formatted list of
the genes that meet the user-specified threshold for
inter-gene influence. This list is passed to one of our
third-party software components, namely, GraphViz[1]. GraphViz processes the list and
returns to ArrayViz a specially coded layout map for our
graphical representation of the GRN. This step is
completely abstracted from the user.
Step 5: Layout Visualized by jGraph
Once ArrayViz has completed all of its internal
computation based on user-specified preferences, and once
GraphViz has given us a coded layout, the final output is
generated by jGraph software[8] in
the form of an undirected graph. The graph is a
representation of a GRN, with each vertex representing a
particular gene in the data set, and edges connecting gene
pairs that directly influence each other. The edges are
also colored and scaled to represent the sign and
magnitude of the influence. The number of vertices
displayed depends on a threshold selected by the user
after being provided with several statistics on the data
set. In Figure 3, an ArrayViz-generated GRN is shown for a
melanoma data set originally containing 587 genes.
Figure 3: GRN for melanoma data set.
Step 6: Graphical Interaction in
ArrayViz
In the end, the biologist is face-to-face with a graphical
abstraction of patients, doctors, scientists, laboratory
work, microarrays, spreadsheets, mathematics and
statistics, and computer science. This picture, however,
is more of a gateway than an end result. If the biologist
sees, for instance, a vertex in the graph that appears to
be "holding it all together," he/she might be
interested in which gene that vertex represents, and might
want to obtain more information. At this point, ArrayViz
lets the user discover which gene the vertex represents by
simply placing the mouse over it. In the next release of
the software, right-clicking a gene will pull up a context
menu linking the user to various biological databases
where further information on that gene and its functions,
as well as countless papers and other reference materials,
can be found with relative ease.
Conclusions
At this point I have used the example of ArrayViz to demonstrate how the right mentoring and training can allow undergraduates with a computer science or computer systems engineering background to contribute to the field of bioinformatics. I have traced multidisciplinary training through each of the fields needed to begin the research, and have shown how that training was put into action with the example of the ArrayViz software. Hopefully, however, I have also shown the scope of the bioinformatics field, and that GRNs represent only one area in this vast and ever-evolving discipline. Perhaps as more and more young researchers realize that they possess much of the required knowledge and many of the required skills to work in this field, many of them will take the initiative in producing new knowledge, which will in turn contribute to the greater good.
Acknowledgments
None of the learning and progress described in this article would have been possible without the patient mentoring received from Dr. Seungchan Kim. His kind, well-timed advice and his consistent testing of my limits has pushed me to places I never knew I could go. Also, the Fulton Undergraduate Research Initiative at the Ira A. Fulton School of Engineering at Arizona State University has been instrumental in facilitating and funding the research project described herein. To that program, its founders, and its staff, I am eternally grateful. A heartfelt thank you is also directed to Mr. Ira A. Fulton for his selfless philanthropy, the Department of Computer Science and Engineering at ASU, as well as the Translational Genomics Research Institute in Phoenix.
References
- 1
- AT&T Labs Contributors. AT&T Labs' GraphViz. 2006. http://www.graphviz.org/ (accessed 07 July 2006).
- 2
- Biodesign Contributors. The Biodesign Institute at Arizona State University. 2006. http://www.biodesign.asu.edu/ (accessed 07 July 2006).
- 3
- BMI Contributors. Department of Biomedical Informatics at ASU. http://bmi.asu.edu (accessed 07 July 2006).
- 4
- Draghici, Sorin. 2003. Data Analysis Tools for DNA Microarrays. Boca Raton: Chapman & Hall/CRC.
- 5
- FURI Contributors. Fulton Undergraduate Research Initiative. 2006. http://www.fulton.asu.edu/furi (accessed 07 July 2006).
- 6
- GEO Contributors. Gene Expression Omnibus. 2006. http://www.ncbi.nlm.nih.gov/geo/ (accessed 07 July 2006).
- 7
- Intel Contributors. TGen/ASU Advances Genomic Research with High-Performance Cluster based on Intel Processors and Linux. Intel Business Computing Case Study. 2003. (accessed March 25, 2006).
- 8
- jGraph Contributors. jGraph Ltd., jGraph Software. 2006. http://www.jgraph.com/ (accessed 07 July 2006).
- 9
- TGen Contributors. Translational Genomics Research Institute. 2006. http://www.tgen.org (accessed 07 July 2006).
- 10
- Verdicchio, Mike. ArrayViz Software. 2006. http://www.public.asu.edu/~mverdic/arrayviz/ (accessed 07 July 2006).
- 11
- Wikipedia Contributors. Bioinformatics. 2006. http://en.wikipedia.org/w/index.php?title=Bioinformatics&oldid=42672144 (accessed March 9, 2006).
Biography
Mike Verdicchio graduated from the Ira A. Fulton School of Engineering at Arizona State University in May of 2006 with a degree in Computer Systems Engineering. As an undergraduate student under the mentorship of Seungchan Kim, PhD, Verdicchio performed bioinformatics research under the Fulton Undergraduate Research Initiative at Arizona State.