Entropy based disease classification of proteomic mass spectrometry data of the human serum by a support vector machine

Disease diagnostics using proteomic patterns is a new platform that is developed to detect early-stage cancer. Proteomic pattern analysis uses the overall pattern to diagnose disease states without the need to identify the components within the pattern. The patterns are generated from mass spectrometry (MS) data, and an algorithm is developed to decipher the patterns within the mass spectrometry data to discriminate between serum taken from healthy and cancer-affected individuals. There is need for cancer biomarkers with more accurate diagnostic capability. Use of MS is such a technique. Mass spectrometry data of the human serum consist of intensities of various ions present in the sample. A typical sample can have about 15000 different ions present. A very important question then is which ions are the best classifiers. We have used an information-theoretical concept, information gain, to measure how well a given attribute separates the training examples according to the target classification. The method measures the drop in the entropy of the system caused by selecting a particular attribute. The lower the drop, the better the attribute. Our algorithm first selects the attributes with highest information gain and then classifies the diseased and healthy data based on these attributes using support vector machines (SVM). The method achieves very strong performance.