DNA Microarray Technology, Data Mining Help Researchers Differentiate Among Patients with Multiple Sclerosis

BUFFALO, N.Y. -- A multidisciplinary team of pharmaceutics and computer-science researchers at the University at Buffalo, one of very few teams in the nation applying DNA microarray technology in studies of multiple sclerosis (MS), has developed a method of interpreting the massive amount of information that results from such experiments.

The application of this method to genomic data could help researchers in the UB School of Pharmacy and Pharmaceutical Sciences, the School of Medicine and Biomedical Sciences and Buffalo General Hospital to develop and ultimately predict the best treatment strategies for MS patients based on levels of gene expression, the process by which a gene's coded information is converted into, or expressed, as proteins in cells.

Results of the work -- which involves using cluster analysis to differentiate healthy controls, MS patients and MS patients being treated with inteferon-beta based on microarray data -- was presented recently at the First Society for Industrial and Applied Mathematics Conference on Data Mining in Chicago.

The team is led by Aidong Zhang, Ph.D., UB associate professor of computer science and engineering, and Murali Ramanathan, Ph.D., UB associate professor of pharmaceutical sciences.

"MS is a complex disease and multiple genes contribute to it," Ramanathan explained. In patients receiving interferon-beta, the first treatment shown to be effective in delaying symptoms in high-risk MS patients, gene expression is affected by both the disease and the treatment. The use of interferon-beta, now marketed under the brand name Avonex(r) by Biogen, was pioneered by Lawrence D. Jacobs, M.D., UB professor of neurology and Irvin and Rosemary Smith Chair in Neurology in the UB medical school.

Microarray technology, Ramanathan explained, is responsible for bringing these kinds of explorations into the realm of possibility because it can measure the expression of thousands of individual genes caused by drugs or disease.

"Before we had this technology, we couldn't even have considered doing experiments of this magnitude," he said.

"Now we can get data on thousands of genes just from a single patient blood sample," he said. "It's parallel processing in biological terms."

But in any given gene expression profile that results from microarray measurements, Ramanathan explained, the majority of the genes being scanned do not provide useful information; the challenge lies in finding those that do.

"We need to find which of these genes change and which do not, so that we then can determine which changes reliably are associated with disease effects and which with treatment effects," he said.

That is a job for computer scientists, or more precisely, data-mining specialists.

"Data is of little use without intelligence," said Zhang.

She explained that data mining is used in a broad spectrum of organizations, ranging from companies such as banks, where it is used to model and predict credit fraud, to pharmaceutical firms, which are beginning to use it to detect potentially effective compounds as the basis for new drugs.

"Data mining is the process of extracting valid, previously unknown and ultimately comprehensible information from large databases and using it to make crucial decisions," she said. "It permits organizations to make the most effective use of data that they have gathered."

The focus of Zhang's group is to look at automated detection of patterns and to devise rules for interpreting that data, using various types of analyses, including a technique called cluster analysis.

Using a method called "maximum entropy," Zhang and her colleagues developed a clustering algorithm to classify the populations in Ramanathan's study as either a healthy control, an untreated MS patient or an MS patient being treated with interferon-beta.

They do it by converting the numeric information generated in a gene expression profile into a feature model that reveals a pattern that may be different or worth highlighting.

The data in those features then may be measured, revealing potentially relevant differing levels of gene expression.

"It would be impossible for me to get this information without the good work of my colleagues in computer science," said Ramanathan, who, in addition to Zhang, works with Raj Acharya, Ph.D., chair and professor in the Department of Computer Science and Engineering affiliated with UB's College of Arts and Sciences and School of Engineering and Applied Sciences, who is a specialist in pattern recognition.

Co-authors on the paper were Shumei Jiang, Chun Tang and Li Zhang, all doctoral students in the UB Department of Computer Science and Engineering.