CUE Home >> Features >> Patterns and Predictions—Computational Methods Shed Light on Genomes

CUE 2006

Computer Science: Patterns and Predictions—Computational Methods Shed Light on Genomes

As a mathematician and computer scientist, Debra Goldberg sometimes sees patterns where biologists can't—and that can make a big difference in the decoding of complex genomes.

Although we now know the genome sequence for humans and many other organisms, Goldberg says that's just the beginning. She likens the massive amounts of data coming from genome projects to a giant jigsaw puzzle, one that can be solved by noting certain patterns. Each discovered pattern helps solve another portion of the puzzle.

By developing combinatorial and graph algorithms to solve problems in modern genomics, Goldberg is helping to find some of these patterns.

After earning an undergraduate degree in biology, Goldberg moved into computer science and spent 10 years in the software industry. She then went on to study computer science and applied mathematics at the graduate level, receiving her PhD from Cornell University where she specialized in computational molecular biology. She joined the computer science department at CU-Boulder in 2006 after spending four years as a research fellow at Harvard Medical School.

"I have a particular interest in helping biologists determine the function of uncharacterized genes or proteins," Goldberg says. "Understanding how proteins function is essential if we hope to gain a systems-level understanding of any organism, and will provide critical new targets to stimulate the development of new medicines."

One way to predict a gene function is to relate an uncharacterized gene in one organism to a known gene in another organism. While Goldberg was a graduate student, she developed a computer-generated comparison of rice and corn in collaboration with a plant biologist, which suggested relationships between the two species not previously known. Such information can be very useful in plant breeding as experts work to introduce greater disease resistance to one species based on its genetic similarities to another.

The dynamic programming algorithm Goldberg and her colleagues developed is similar to that used for analyzing human and programming languages. It produces quick results by relating results to those already found, in effect considering a wide range of possibilities without actually going through every comparison. The algorithm was used to further develop a comparison of mice and human genomes, which could be useful in the understanding of genes associated with human diseases.

Goldberg also examined patterns in the interconnections of protein networks in organisms such as yeast and worms. A network is a model of objects and their associations that is used to study complex systems as diverse as social networks, electric power grids, and the World Wide Web. In a network, objects are represented by a point called a node or vertex, and associations are represented as a line or edge between two object nodes. In the networks that Goldberg studies, the nodes represent proteins, and the edges may represent a variety of relationships, such as the physical interaction or co-expression (synthesis at the same times and conditions) of two proteins. The function of a previously uncharacterized protein can be predicted based on its neighbors in the network and the patterns of edges in the neighborhood around the protein.

Such system-level approaches have great potential for increasing our understanding of biological organisms, while also speeding up the characterization of large numbers of genes and proteins through high-throughput analysis.

For more information, visit www.cs.colorado.edu/~debra
Department web site: www.cs.colorado.edu