Statistical Treatment of Proteomic Imaging Mass Spectrometry Data

Abstract

Proteomic imaging mass spectrometry is an emerging field, and produces large amounts of high-dimensional data. We propose approaches to extracting useful information from these data - two of particular note. The Difference in Proportions of Occurrence Statistic (DIPPS) applies to binary data and leads to easily interpretable maps useful for exploratory analyses and automated generation of feature lists that can be used to standardise comparisons between datasets. The second approach, based on Canonical Correlation Analysis (CCA), reduces the high-dimensional data to features strongly related to classes and leads to good classification. Applications to cancer data show the success of these approaches.