Multi-class alignment of LC-MS data using probabilistic-based mixture regression models

Abstract

In this paper, a framework of probabilistic-based mixture regression models (PMRM) is presented for multi-class alignment of liquid chromatography-mass spectrometry (LC-MS) data. The proposed framework performs the alignment in both time and measurement spaces of the LC-MS spectra. The expectation maximization (EM) algorithm is used to estimate the joint parameters of spline-based mixture regression models and prior transformation densities. The latter are incorporated to account for variability in time and measurement spaces of the data. As a proof of concept, the proposed method is applied to align a single-class replicate LC-MS spectra generated from proteins of lysed E.coli cells. Its performance is compared with the dynamic time warping (DTW) and continuous profile model (CPM) approaches.

Current Technological Challenges in Biomarker Discovery and Validation

“…A large number of different methods have been developed for the automated correction of non-linear retention time shifts. [116][117][118][119][120][121][122][123][124][125][126][127][128] the methods differ principally in that some use the raw data 120,124,127 while others work with peak lists. 122,123,128 Another principal difference is that some methods use the global method (they use the entire chromatogram for alignment in each step of the alignment procedure) while others try to find the best time alignment path between two chromatograms by means of dynamic programming using point-by-point or segmentwise optimization procedures.…”

Statistical Analysis of Chemical Sensor Data

“…Coombes et al (2005); Lin et al (2005); Morris et al (2005)). While analogous approaches can likewise be applied to two-dimensional data, other methods have also been suggested, which include filtering in the wavelet domain , and asymmetric least squares splines regression (Befekadu et al;.…”

LC-MS Data Analysis for Differential Protein Expression Detection

“…Moreover, the framework lends itself to an expectation-maximization (EM) algorithm with the following features (i) the explicit use of transformation priors for modeling the ion abundance (peak intensity) variability in both RT and m/z dimensions of the data, (ii) the use of a probabilistic metric that allows estimation of the distance among multiple LC-MS data instead of computing pair-wise distances, and (iii) the ability to extend the method for alignment and normalization of LC-MS data involving multiple groups. We demonstrated that analysis of LC-MS data via PMRM has the potential to address critical concerns such as unequal intervals across multiple runs and misalignment both in time and measurement space (18,19). We assume that the observed dataset D representing multiple groups is generated with the following three features (i) an individual is randomly drawn from a population of M objects (i.e., the dataset D); (ii) the individual is assigned to the kth group with probability y .…”

“…For example, a recently introduced method called continuous profile model (CPM) has been applied for alignment and normalization of continuous time-series data and for detection of differences in multiple LC-MS data (6,17). Similarly, we developed a probabilistic mixture regression model (PMRM) for global alignment of LC-MS data (18,19). We approach the problem of LC-MS data alignment with an ultimate goal of detecting differences among groups of LC-MS runs.…”