UVM Theses and Dissertations
Format:
Print
Author:
Compton, Duane Charles
Dept./Program:
Computer Science
Year:
2008
Degree:
MS
Abstract:
TWiGS (two-dimensional wavelet transform with generalized cross validation and soft thresholding) is a novel algorithm for denoising liquid chromatography-mass spectrometry (LC-MS) data for use in "shot-gun" proteomics. Proteomics, the study of all proteins in an organism, is an emerging field that has already proven successful for drug and disease discovery in humans, and LC-MS is its most promising tool. In simple terms, liquid chromatography (LC) can be thought of as a chemical sieve that separates compounds based on their physical and chemical interactions with a tightly packed silica column. Mass spectrometry (MS) is used to determine the mass of an ionized compound that has eluted from the LC column. The resulting data is presented as a two-dimensional digital signal (the LC transit time versus the mass-to-charge ratio, m/z). Each chemical species contributes one or more asymmetric peaks.
There are a number of constraints that limit the effectiveness of LC-MS for shot-gun proteomics, where the chemical signals are typically weak, and data sets are too large to separate the signal from noise by hand. Traditional LC and MS noise removal focus on only one domain of the two-dimensional data set, and signal processing is largely limited to time-series analysis. Most algorithms suffer greatly from a researcher driven bias, making the algorithms and results irreproducible and unusable by other laboratories. We thus introduce a new algorithm, TWiGS, that removes electrical (additive white) and additive chemical noise from LC-MS data sets.
TWiGS is developed to be a true two-dimensional algorithm, which operates in the timefrequency domain, and minimizes researcher bias. It is based on the discrete wavelet transform (DWT), which allows for fast and reproducible data analysis. The separable two-dimensional DWT is paired with generalized cross validation (GCV) and soft thresholding. The Haar, Coiflet-6 (C6), Daubechies-4 (D4) and the number of decomposition levels are determined based on observed experimental results. Using a synthetic LC-MS data model, TWiGS accurately retains key characteristics of the peaks in both the time and m/z domain, and can detect peals from noise of the same intensity. TWiGS is applied to angiotensin I and II samples run on a LC-ESI-TOF-MS (Liquid-Chromatographjr-Electrospray-Ionization) to demonstrate its utility for the detection of low-lying peaks obscured by white noise.
Synthetic and experimental data analysis show that TWiGS can resolve low-lying peaks, while retaining key data parameters. As the amount of noise created from electrical signal amplification used in MS to determine a molecules mass-to-charge ration (m/z) is unlikely to be reduced in coming years, TWiGS offers a solution to researchers who wish to gain the most information from their data as possible. Reproducibility is a built in feature of TWIGS by means of automatic threshold selection, making it a crucial launching point for higher level processing algorithms. TWiGS is extensible to any LC-MS based analysis, making it useful to a plethora of research fields, however in this work we focuses on TWiGS role in shot-gun proteomic research.
There are a number of constraints that limit the effectiveness of LC-MS for shot-gun proteomics, where the chemical signals are typically weak, and data sets are too large to separate the signal from noise by hand. Traditional LC and MS noise removal focus on only one domain of the two-dimensional data set, and signal processing is largely limited to time-series analysis. Most algorithms suffer greatly from a researcher driven bias, making the algorithms and results irreproducible and unusable by other laboratories. We thus introduce a new algorithm, TWiGS, that removes electrical (additive white) and additive chemical noise from LC-MS data sets.
TWiGS is developed to be a true two-dimensional algorithm, which operates in the timefrequency domain, and minimizes researcher bias. It is based on the discrete wavelet transform (DWT), which allows for fast and reproducible data analysis. The separable two-dimensional DWT is paired with generalized cross validation (GCV) and soft thresholding. The Haar, Coiflet-6 (C6), Daubechies-4 (D4) and the number of decomposition levels are determined based on observed experimental results. Using a synthetic LC-MS data model, TWiGS accurately retains key characteristics of the peaks in both the time and m/z domain, and can detect peals from noise of the same intensity. TWiGS is applied to angiotensin I and II samples run on a LC-ESI-TOF-MS (Liquid-Chromatographjr-Electrospray-Ionization) to demonstrate its utility for the detection of low-lying peaks obscured by white noise.
Synthetic and experimental data analysis show that TWiGS can resolve low-lying peaks, while retaining key data parameters. As the amount of noise created from electrical signal amplification used in MS to determine a molecules mass-to-charge ration (m/z) is unlikely to be reduced in coming years, TWiGS offers a solution to researchers who wish to gain the most information from their data as possible. Reproducibility is a built in feature of TWIGS by means of automatic threshold selection, making it a crucial launching point for higher level processing algorithms. TWiGS is extensible to any LC-MS based analysis, making it useful to a plethora of research fields, however in this work we focuses on TWiGS role in shot-gun proteomic research.