of breast cancer as compared with ovarian cancer. However, to explore the possibility of serum proteomic profiling as a supplement to biopsy, an initial study was performed on serum samples from 317 patients who received breast biopsies. A training set, consisting of sera from 43 patients with benign lesions and 58 with breast cancer, was used to identify a discriminating pattern, which, when blind-tested on the remaining 216 samples, detected breast cancer with 90 percent sensitivity and benign lesions with 71 percent specificity.36 These results suggest the feasibility of using this methodology as a supplement to biopsy, as well as the need for significant improvement in the accuracy of proteomic diagnosis before it could be substituted for the actual results of biopsy. However, a serum proteomic test will only reveal certain biological characteristics of a tumor; other characteristics such as the size, shape, and location of the tumor remain unknown when using only a serum test. Therefore, in using this technique, lesions will most likely still have to be imaged by modalities such as mammogram, ultrasound, and magnetic resonance imaging (MRI), for effective treatment.
In addition to previously described technical challenges to the clinical adaptation of DNA and protein microarrays and of serum proteomic profiling, these high-throughput, biologically based technologies face several barriers to development for the detection, diagnosis, or monitoring of cancer. Two largely unmet requirements stand out: to validate a strong and reliable link between profile characteristics and clinical outcomes and to create reliable, cost-effective profiling methods that can be performed in the clinical setting.52
The accurate analysis and correct interpretation of data from high-throughput experiments, key factors in establishing the clinical significance of molecular profiles, are far from ensured. Many sources of noise can obscure the results of these experiments. Results generated by DNA microarrays, for example, may be influenced by methods of sample storage, preparation, and labeling; by spot location on the array; or by imperfections in the array itself. These problems were clearly illustrated in a recent study in which samples from the same tissue, analyzed with different DNA microarray technologies (cDNA versus oligonucleotide), produced different gene expression profiles.32 In the case of serum proteomic profiling, where the identity of the specific proteins is unknown, minor differences in specimen procurement and subsequent handling may introduce systematic but undetectable biases into profiles.
Thus it is perhaps not surprising that statisticians have warned of significant potential for error in the analysis of voluminous genomic and