Searching through biological images
Researchers at the Arizona State University (ASU) are working on software tools to analyze databases of biological images. One of these projects is using machine learning technology to compare the expression patterns captured in the images. So far, the software was used to explore a database of embryonic fruit flies images to see if the genes share the same spatial patterns. This would indicate that these genes also share similar functions. The goal of the developers is to build a tool able to search biological image databases as fast as Internet search engines are doing.
As an example, you can see on the left a collage of fruit fly gene expression images. "The proper development of each football-shaped fly embryo depends on the coordinated expression of thousands of genes. By studying the expression pattern of single genes, typically displayed in wide bands or narrow striped patterns, scientists can gain insight into the control and regulation of large genetic networks. Similar gene networks are found throughout biology, and break downs in these processes may result in birth defects, heart disease, cancer and aging. (Credit for image and caption: Biodesign News at ASU)
This project is led by Jieping Ye, an assistant professor of computer science also working for the Biodesign Institute's Center For Evolutionary Functional Genomics at ASU. Here are some more details about his project.
Ye, a computer scientist, is using a $583,603 grant to develop a computational framework for analyzing biological images. His system uses a technology called "machine learning," a technique routinely used to recognize faces and to thwart credit card fraud. In this case, the "faces" are a large collection of Drosophila (fruit fly) embryonic images obtained through the Berkeley Drosophila Genome Project.
Each image in the collection represents a different stage in time of an embryo's development and a specific pattern of gene expression. "The key in this project is to be able to compare the expression patterns captured in the images," Ye says. "If the images share similar spatial expression patterns, then the genes may share the same function."
But even if Ye can successfully build a good search engine for databases of biological images, there is still the -- not so trivial -- problem of preparing the data to be used.
Another challenge in the project is to prepare all of the biological images, so that they can be compared with one another. This involves a painstaking process in which each image is manually adjusted to the same size, shape and orientation. So far, the research group has combed through more than 45,000 images, which is about half of the images available.
If you want more information about this project, you can read a paper presented last year at the Computational Systems Bioinformatics Conference (CSB), "Classification of Drosophila embryonic developmental stage range based on gene expression pattern images" (PDF format, 6 pages, 2.52 MB).
Sources: ASU Insight, May 1, 2007; and various websites
You'll find related stories by following the links below.