Developments to Watch
HUNTING THROUGH THE `GARBAGE' FOR DNA
SCIENTISTS HAVE FOUND The makeup and function of thousands of genes in creatures from bacteria to rats. And because evolution tends to preserve traits that work, many of these genes have human cousins. But identifying the human version isn't easy. Genes aren't an unbroken stretch of DNA text, like a movie. They're more like TV shows, with each segment of information surrounded by ads--what researchers call genetic garbage. Both the length of segments (called exons) and the amount of "trash" can vary dramatically from species to species.
Existing software programs already find human genes by looking for DNA sequences that are similar to those in animal genes. But Russian mathematicians, including one working at the University of Southern California, have devised a potentially better way. Their approach divides up strings of animal DNA many different ways, then looks for the combination of pages and surrounding garbage that produces the best fit with the unknown DNA. In some tests, the method's accuracy rate is near 100%.EDITED BY PETER COY By John CareyReturn to top
SO MUCH FOR THE PRINCIPLE OF SIMPLICITY
THE SIMPLEST EXPLANATION is always best, right? Wrong. In fact, research by Geoffrey I. Webb, a computer scientist at Australia's Deakin University, found major flaws in the "Occam's razor" principle, which states that explanations should be pared down to the simplest possible set of factors.
This has major implications for many software tools, especially data-mining programs designed to search databases and extract hidden relationships among facts, such as how customer service affects brand loyalty. Data mining is spreading rapidly as companies learn the extent of untapped information they already have--and how valuable it can be for marketing and competitiveness. But most such programs don't question the validity of Occam's razor, credited to William of Occam (or Ockham), a 14th-century English philosopher. As a result, says Webb, users are probably missing much of the hidden knowledge--or worse, drawing misleading conclusions.
To put Occam's razor to the test, Webb modified the most popular machine-learning program, called C4.5. Its learning is guided by both Occam's razor and a related principle known as assumption of similarity, which compares a problem to known solutions. C4.5 gives precedence to Occam's razor, but Webb changed that and made the similarity assumption, which gives more complex results, the preferred approach. Then he tested it for various decision-making tasks where the best outcomes were known. The "decision trees" developed by the similarity assumption technique were significantly better, says Webb--so much so that he regards Occam's razor as worse than dull: "It is truly disposable."EDITED BY PETER COY By Otis PortReturn to top
NOW, WORRY-FREE WATER?
AN OUTBREAK OF CRYPTOSPORIDIUM IN MILWAUKEE'S Drinking water killed about 100 people in 1993. A sand-and-charcoal filter couldn't cope with massive contamination by the parasite. Partly in response to that disaster, Congress passed the Safe Drinking Water Act, which requires utilities to upgrade their treatment facilities.
Sionix Corp., a San Diego startup, says it can help utilities meet the new standards with a filter designed to cleanse water of particles as small as 2 microns in diameter. That's half the size of cryptosporidium egg cases. First, microscopic bubbles injected into the water attach to contaminants and float them to the surface, where they can be skimmed off. Next, ozone mixed into the water kills germs. Finally, the water passes through a fine stainless-steel mesh. Ordinarily such a mesh would become clogged almost instantly. But Sionix recently received a U.S. patent on a way to swirl the water so it scours the mesh, reducing the need for backflushing.EDITED BY PETER COYReturn to top