Scientists at the State University of New York, Stony Brook say they have developed a new algorithm that identifies with 100 percent sensitivity and 100 percent specificity cases of ovarian cancer based on mass spec-produced patterns of proteins in serum. The study, published in the Dec. 9 issue of Proceedings of the National Academy of Sciences, adds another tool to the 2-year-old project led by Emanuel Petricoin and Lance Liotta to develop a pattern-based ovarian cancer diagnostic test, according to Petricoin and Wei Zhu, assistant professor of applied mathematics and statistics at SUNY Stony Brook, who led the new effort.
“This just lends support to what we initially said two years ago,” Petricoin said. “The more people that can reproduce what we do, the more people think it’s real.”
Zhu developed the algorithm using serum proteomic data sets available online from the NIH and FDA Clinical Proteomics Program Databank. The algorithm determines the patterns that are indicative of cancer based on pre-selection of differentially expressed markers and scanning of entire spectra — rather than the “random window” approach used by Correlogic’s Proteome Quest algorithm, the mass spec-based ovarian cancer diagnostic algorithm being used in partnership with Petricoin. Correlogic has already licensed its algorithm test, which has achieved 100 percent sensitivity and 97 percent specificity according to Zhu, for selected home brew use (see PM 7-18-03).
Although the data that Zhu used came from earlier sets in which Petricoin’s NCI-FDA group used SELDI-TOF to create the spectra, Zhu told ProteoMonitor’s sister publication BioInform that she has arranged to receive newer data from the ProteinChip/ABI Q-STAR higher resolution setup that the NCI-FDA group is using now.
Petricoin downplayed the difference between the 100 percent sensitivity and 100 percent specifi-city that Zhu’s group achieved and the 100 percent/97 percent numbers that Correlogic’s algorithm achieved, calling it a “hair-splitting question.” Still, he said that he is in talks with Zhu to see how the new algorithm can be incorporated into versions that the NCI-FDA and its collaborators are already using.
“Scientifically, we’re interested in using several different algorithms at once so we can find features that are consistently found,” he said. “We actually have a number of different algorithms that we use and different collaborators’ algorithms that we use, and we’re looking for concordance.”
Zhu agreed in principle that combining several algorithms to develop the best possible test was the way to go, but argued that when it comes to ovarian cancer, the difference between 100 percent and 97 percent specificity is actually significant. “If we want to use this screening test with 97 percent specificity for mass screening, what would happen? For every correctly diagnosed cancer case, there would be 75 false positives identified at the same time,” she said. “So for ovarian cancer screening, the specificity must be near 100 percent. It’s like an HIV test — you don’t want an HIV test with 97 percent specificity.”
Meanwhile, the first version of an ovarian cancer diagnostic test is expected to hit the market in the first quarter of 2004 (see PM 11-14-03).