Abstract

We present model-based inference for proteomic peak identification and quantification from mass spectroscopy data, focusing on nonparametric Bayesian models. Using experimental data generated from MALDI-TOF mass spectroscopy (matrix-assisted laser desorption ionization time-of-flight) we model observed intensities in spectra with a hierarchical nonparametric model for expected intensity as a function of time-of-flight. We express the unknown intensity function as a sum of kernel functions, a natural choice of basis functions for modeling spectral peaks. We discuss how to place prior distributions on the unknown functions using Lévy random fields and describe posterior inference via a reversible jump Markov chain Monte Carlo algorithm.

Introduction

The advent of matrix-assisted laser desorption/ionization such time-of-flight (MALDI-TOF) mass spectroscopy and related SELDI-TOF (surface enhanced laser desorption/ionization) allows the simultaneous assay of thousands of proteins, and has transformed research in protein regulation underlying complex physiological processes. This technology provides the means to detect large proteins in a range of biological samples, from serum and urine to complex tissues, such as tumors and muscle. With appropriate statistical analysis, one may explore patterns of protein expression on a large scale in high-throughput studies without the need for prior knowledge of which proteins may be present (Baldwin et al., 2001; Diamandis, 2003; Martin and Nelson, 2001; Petricoin and Liotta, 2003; Petricoin et al., 2002). As such, it becomes a discovery tool, identifying proteins and pathways that are linked to a biological process. In applications, tens to thousands of spectra may be collected, leading to massive volumes of data. Each spectrum contains on the order of tens of thousands of intensity measurements, with an unknown number of peaks representing proteins of specific mass-to-charge ratios.