Bayesian Nonparametric Approach for Analyzing Mass Spectrometry Data

• Activity Number: 46
• Type: Topic Contributed
• Date/Time: Sunday, August 8, 2004 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Leanna L. House*+ and Merlise Clyde
• Companies: Duke University and Duke University
• Keywords: Bayesian ; overcomplete ; nonparametric ; gamma-convolution model ; proteomics ; MALDI-TOF MS

Abstract

Motivated by the science of expression proteomics, we develop a Bayesian nonparametric approach for predicting disease status from high-dimensional proteomic profiles assessed by MALDI-TOF Mass Spectrometry (MS). We develop a nonparametric model for each individual's spectrum through a process convolution, using marked gamma processes. The gamma-convolution model has several desirable features: ensures non-negativity of the modeled protein abundance, allows for shifting of peak locations in the observed spectra, and captures dependencies. Locations of jumps in the gamma process can be used to identify peaks, and associated marks used to identify which peaks are associated with differential protein expression, and predictive of disease status. The gamma-convolution model can be represented as an overcomplete kernel regression model with a Poisson number of components. Using this representation, we capitalize on a reversible jump Markov chain Monte Carlo algorithm to sample from the posterior distribution. Using Bayesian model averaging, we make probabilistic statements concerning the predictions of patient disease status.