In response to recent suggestions that the patterns found in their 2002 Lancet study may have been due to noise, bias, or other experimental error, Emanuel Petricoin and Lance Liotta struck back last week with a published rebuttal that sought to refute this theory as well as address other common criticisms of their work.
Published in BMC Bioinformatics (click here for article), Petricoin and Liotta’s rebuttal focused on a June 2003 paper in BMC — written by James Sorace and Min Zhan of the University of Maryland — that asserted that markers in the low m/z range of the publicly available dataset from their initial SELDI-TOF experiments could be attributed to experimental bias.
A similar critical analysis published in January in Bioinformatics by a team from the M.D. Anderson Cancer Center also argued that some of the pattern elements could be artifacts (see PM 2-13-04).
“We respectfully challenge these broad conclusions as judgmentally biased and scientifically unfounded, and propose that the authors made assumptions that are incorrect considering the physical principles and standard calibration methods of mass spectrometry,” Petricoin and Liotta’s rebuttal read.
In their rebuttal, the two scientists said that they had accounted for potential biases by treating all samples equally and in a blinded manner, and that events found in the low molecular weight region could not necessarily be written off as artifacts. They challenged the expertise of Sorace and Zhan in the area of mass spectrometry and questioned their decision not to contact Petricoin and Liotta’s group before publishing their critique.
“The aforementioned issues highlight the growing need for the editorial process to more thoughtfully consider the possibility that neither the ‘consumer’ author nor the reviewers have all of the necessary information about the posted data sets before a final conclusion is rendered,” Petricoin and Liotta wrote in the article.
Petricoin and Liotta took the opportunity to address other criticisms of their work as well — in particular their decision not to publish identifications of the elements of the originally published patterns. In both the published rebuttal and in a conversation with ProteoMonitor last month, the scientists vehemently defended their decision not to publish identifications of the elements of the original patterns, as well as their assertion that patterns may be useful without a formal identification.
“We have candidate markers that we have identified but until we are absolutely sure they are the right ones we’re not going to publish them because we want to be as rigorous and correct as possible,” Liotta told ProteoMonitor last month. He said that his group did a “rational triaging” of the many pattern elements they found to decide which ones to spend time identifying. He and Petricoin also argued in the published rebuttal that only those patterns that have been validated extensively should be put through the “labor intensive” sequencing process.
“Our quandary is that we have so much content that people think wrongly that the pattern of ions we published in the Lancet paper is the only thing we found — that’s just one of dozens and dozens and dozens of patterns that all did equally well,” Petricoin added. He complained both in the conversation and in the written rebuttal that most scientists who had critiqued the data had so far ignored the higher resolution data that his group produced more recently using an ABI Q-STAR, although this data has been available on the FDA-NCI Clinical Proteomics website for the past eight months.
Petricoin also offered an explanation for why he felt protein identification should not necessarily be required before a pattern test could be offered as a diagnostic — an argument that is becoming increasingly rare as more and more scientists in biomarker discovery insist upon marker identification (see PM 2-13-04). “People ignore the fact that analytes like PSA and CA125 were used clinically for many years without identifying the proteins,” Petricoin told ProteoMonitor. He compared the use of a band on an immunoassay — which for many years was used to identify the presence of CA125, for example — to a peak on a mass spec. “The precedent for using and measuring entities that are not identified has already been set,” he said.
In the rebuttal, Petricoin and Liotta took this argument further, suggesting that it in fact would be irresponsible to wait for identifications before running with a potential diagnostic. “The decision not to explore promising approaches that offer clinical utility today would be scientifically unjustified, and morally wrong,” they wrote.