Automatic selection of preprocessing methods for improving predictions on mass spectrometry protein profiles

Mass spectrometry proteomic profiling has potential to be a useful clinical screening tool. One obstacle is providing a standardized method for preprocessing the noisy raw data. We have developed a system for automatically determining a set of preprocessing methods among several candidates. Our system's automated nature relieves the analyst of the need to be knowledgeable about which methods to use on any given dataset. Each stage of preprocessing is approached with many competing methods. We introduce metrics which are used to balance each method's attempts to correct noise versus preserving valuable discriminative information. We demonstrate the benefit of our preprocessing system on several SELDI and MALDI mass spectrometry datasets. Downstream classification is improved when using our system to preprocess the data.