Canonical Correlation, an Approximation, and the Prediction of Protein Abundance

This paper addresses a central problem of Bioinformatics and Proteomics: estimating the amounts of each of the thousands of proteins in a cell culture or tissue sample. Although laboratory methods involving isotopes have been developed for this problem, we seek a simpler method, one that uses fewer laboratory procedures. Specifically, our aim is to use data-mining methods to infer protein levels from the relatively cheap and abundant data available from high-throughput tandem mass spectrometry (MS/MS). In this paper, we develop and evaluate a method for tackling this problem. The method is based on a simple generative model of MS/MS data. We first show how to linearize the model and fit it to data using Canonical Correlation Analysis (CCA). Then, because CCA is computationally expensive for the large datasets we are dealing with, we develop an efficient approximation of CCA, one that exploits the structure of our data. We prove that the method is correct in that it achieves a well-defined optimization criterion. We also evaluate the method on several biological datasets. The datasets themselves were generated by MS/MS experiments performed on various tissue samples taken from Mouse. keywords: Bioinformatics, Proteomics, Data Mining, Machine Learning, Peptides, Tandem Mass Spectrometry.