TOGETHER, gene technology, artificial brain help cancer diagnosis
Thursday, May 31st 2001, 12:00 am
By: News On 6
WASHINGTON (AP) _ By combining gene technology and high-speed computers that learn as they go, scientists have determined a way to tell the difference among several childhood cancers that appear similar.
Being able to differentiate these diseases quickly allows doctors to improve treatment.
In the long run, they hope, it will lead to better ways to battle the diseases.
``Basically, we are merging two technologies,'' explained Dr. Paul Meltzer of the National Human Genome Research Institute, a part of the National Institutes of Health.
The study, reported in the June issue of the journal Nature Medicine, focused on small, round blue-cell tumors of childhood, a group of four types of cancer that appear similar but respond to different treatments. Those tumors include neuroblastoma, rhabdomyosarcoma, non-Hodgkin's lymphoma and the Ewing family of tumors.
``These cancers are difficult to distinguish, ... and currently no single test can precisely distinguish these cancers,'' the scientists reported.
The first step involves microarrays, which are about the size of a microscope slide and contain specific fragments of specially chosen DNA, the genetic building block.
An active gene sends out a chemical signal called RNA, which carries a particular sequence of building blocks that tell what gene it came from.
A tissue sample from the cancer is spread on the microarray, and each RNA string then attaches to a waiting DNA segment designed to attract it.
By detecting how many strings end up at each site on the chip, instruments can tell how active the corresponding gene had been in the tumor. Many strings signify a very active gene. No strings at a particular site means the corresponding gene had been turned off.
This information is then fed into a computer program called an artificial neural network, which has learned which gene segments are most active for differing types of cancer.
``Essentially it's a computer program that does an artificial learning process,'' Meltzer explained. ``The data goes through a training phase, and the computer tries to learn the features of the data that allow it to make a classification.''
Once the neural network had learned to classify the types of cancer, it was tested in 88 different experiments and ``essentially got them all categorized correctly,'' Meltzer said.
``This is something which, in one form or another, is likely to actually be tested in a clinical setting in the near future,'' he said. He wouldn't speculate how soon.
In addition, Meltzer noted, the process allows scientists to learn gene profiles of the various types of cancer.
``When you discover these profiles, or fingerprints or signatures that represent the profile of gene expression for a given cancer, you're also getting lists of genes ... for possible new therapeutic targets,'' he said.
Yudong He and Stephen Friend of Rosetta Impharmatics in Kirkland, Wash., agreed that a benefit of such analysis is the information that could lead eventually to the design of more specific drugs for various tumors.
They were less confident about how soon the tools could come into use. ``We are still a long way from translating the clues provided by DNA microarrays into diagnostic tools,'' the two wrote in a commentary accompanying Meltzer's paper.
``Although microarrays might eventually be the diagnostic method of choice, they are currently too expensive for routine clinical use,'' wrote He and Friend, who were not part of the research team.
Meltzer said the current cost of the tests can range from $200 to $500, but his team worked out the minimum number of genes that need to be tested to differentiate among the four cancers.
They concluded that testing for 96 genes was enough to tell the cancers apart, and Meltzer said a test for that small a number would be much less costly.