Mass spectrometry sensing data for robust cancer classification

Abstract

The high dimensionality of mass spectrometry (MS) data forces classification processes to be computationally intensive while at the same time, a classifier that is accurate, robust, and can be implemented in real time is needed. In this paper, we present an accurate prediction of class content in a sample of compressed MS data while reducing MS data dimensionality. By presenting the MS data through compressive sensing (CS) sampling, the sensing data does not only have a lower dimension than the original data but also it is well be preserved and can be reconstructed. Classification is established using low dimensional MS data leading to faster processes without any loss of accuracy. We are proposing to use a method based on L2 norm and a mixed L2-L1-norms regularization terms in a classification framework capable of solving an overdetermined system but with a dimensionality reduction attribute. Our results show that L2-algorithm with regularization terms has a better performance than standalone L2-algorithm and Q5 under all applicable assumptions. The performance has a throughput proteomic cancer classification with sensitivity, specificity, and positive predictive values of at least 95%.