An MCMC-MRF Algorithm for Incorporating Spatial Information in IMS Proteomic Data Processing

Abstract

It is desirable to not only identify the peaks of the mass spectra but also to study relations among them using the spatial information for the entire imaging mass spectrometry (IMS) data cube. In this paper, we incorporate spatial information in IMS data analysis using Markov random field (MRF) and optimize classification accuracy with Markov chain Monte Carlo (MCMC) sampling. First, we discuss the necessity of incorporating spatial information in IMS data analysis and give a brief introduction to MRF and its background. Then, we develop the MCMC-MRF computation framework using MCMC sampling and the Ising model, which is the simplest MRF, as prior information to optimize IMS data classification accuracy. The method to estimate parameters using training data is also discussed. Finally, we use test data to test the performance of this model under different definitions of neighboring system. The experiment results show that the MCMC-MRF model can improve IMS data classification accuracy effectively, and the more realistically the neighboring system is defined, the better classification result will be.