Diagnoses That Cast A Wider Net

DOCTORS HAVE LONG treated complex diseases such as cancer with "cocktails" of drugs that fight illness on several fronts. But when it comes to diagnostics, few biotech companies have adopted multipronged strategies. Their tests zero in on a single biochemical marker--usually a particular protein the body releases when hit by a disease.

This narrow focus seems odd, because in the course of an illness, tissues undergo dynamic molecular changes that can leave many different chemical trails in the blood. While humans may have trouble making sense of these complicated patterns, computers that run "smart" software called neural networks excel at distilling information from apparent chaos. So, for five years Horus Therapeutic Inc. in Rochester, N.Y., has been "training" neural networks to discover patterns among multiple biomarkers released when a tumor invades healthy tissue. Once the system understands the variables, it can make highly accurate diagnoses based on blood tests.

In clinical tests involving 800 volunteers, a new test for prostate cancer achieved 90% accuracy--far outperforming the popular, single-marker PSA blood test. The Food & Drug Administration is now assessing the results, but Horus isn't waiting to push ahead with similar tests for colon, ovarian, and breast cancer.