Proteomic MS-spectra decomposition into intensity-regions for identifying potential biomarkers and improving discrimination accuracy between normal and cancer tissue spectra

Aim: In this study a new strategy for processing proteomic MS-spectra is presented for (a) the determination of potential meaningful cancer biomarkers (m/z values), extracted from different MS-spectra intensity regions, and (b) reliable and effective separation of normal from cancer tissue MS-spectra. Method: The method starts by MS-spectra signal conditioning (base line subtraction- normalization-smoothing-noise estimation-peak detection-peak alignment). It then continues by automatically breaking down all MS-spectra into common-equidistant intensity regions. Subsequently, most informative features (m/z values), which might constitute potential significant biomarkers, are determined at each intensity region by using a pattern recognition system for discriminating normal from cancer tissue MS-spectra. Finally, considering selected features from all spectral intensity regions, spectra were classified using a multi-classifier scheme, with members the Support Vector Machine, the Probabilistic Neural Network, and the k-Nearest Neighbour classifiers. To ensure robust and reliable estimates to unseen data, the proposed pattern recognition system was evaluated using an external cross validation process. Material: The proposed method was evaluated on two publicly available proteomic datasets, one with ovarian and the other with prostate MS-spectra. Both datasets were analyzed excluding m/z values lower than 1500, which are potentially distorted, in an effort to analyze more accurately the proteomic datasets. Results: The average overall performance of the system in classifying normal from ovarian cancer MS-spectra was 97.2% (employing 22/24/18/17/15 biomarkers at each intensity region, ranked from highest to lowest intensity regions), whereas the accuracy in discriminating spectra with no evidence of prostate disease (PSA4) was 92.5% (employing 8/4/15/12/13 biomarkers at each intensity region, ranked from highest to lowest intensity regions). Conclusion: The proposed method differs from others in two key issues. (a) The methodology, where the concept of focusing interest on peaks from different MS-spectra intensity-regions was introduced, and (b) the accuracy, where, as compared to previous studies that experimented with m/z values above 1500, the proposed system presented the highest classification accuracies. Additionally, a pattern recognition system is proposed that might be of value in the discrimination of normal from c