Quadratic variance models for adaptively preprocessing SELDI-TOF mass spectrometry data
© Emanuele and Gurbaxani; licensee BioMed Central Ltd. 2010
Received: 12 April 2010
Accepted: 13 October 2010
Published: 13 October 2010
Surface enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI) is a proteomics tool for biomarker discovery and other high throughput applications. Previous studies have identified various areas for improvement in preprocessing algorithms used for protein peak detection. Bottom-up approaches to preprocessing that emphasize modeling SELDI data acquisition are promising avenues of research to find the needed improvements in reproducibility.
We studied the properties of the SELDI detector intensity response to matrix only runs. The intensity fluctuations and noise observed can be characterized by a natural exponential family with quadratic variance function (NEF-QVF) class of distributions. These include as special cases many common distributions arising in practice (e.g.- normal, Poisson). Taking this model into account, we present a modified Antoniadis-Sapatinas wavelet denoising algorithm as the core of our preprocessing program, implemented in MATLAB. The proposed preprocessing approach shows superior peak detection sensitivity compared to MassSpecWavelet for false discovery rate (FDR) values less than 25%.
The NEF-QVF detector model requires that certain parameters be measured from matrix only spectra, leaving implications for new experiment design at the trade-off of slightly increased cost. These additional measurements allow our preprocessing program to adapt to changing noise characteristics arising from intralaboratory and across-laboratory factors. With further development, this approach may lead to improved peak prediction reproducibility and nearly automated, high throughput preprocessing of SELDI data.
Mass spectrometry is a promising technology for biomarker discovery . There are a wide variety of mass spectrometers from which one could choose from during the design of a biomarker discovery experiment, reviewed in . Matrix assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS, or just MALDI) can ionize whole proteins intact over a wide range of protein mass values, making it suitable for biomarker discovery in complex media such as blood serum, where both protein concentrations and masses vary greatly . Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF MS, or just SELDI)  is a variant of MALDI that adds an on-chip chromatographic separation step at the front end of the analysis pipeline. This, combined with robot-automated sample preparation, enables SELDI to be high-throughput, an attractive feature for many laboratories. For a recent review of the application of SELDI in the context of biomarker discovery, see .
The typical SELDI work flow involves the collection of samples (e.g.- blood serum) from patients, application of the samples to SELDI ProteinChips® selected for desired physicochemical properties, and analysis in the SELDI mass spectrometer. The raw data must be preprocessed to detect relevant peaks which correspond to proteins in the sample. Typical signal preprocessing steps performed are spectral alignment, denoising/smoothing, peak detection, peak matching, normalization, and quantification (see Figure 1 of ). The preprocessing of the raw SELDI spectra is typically accomplished using one of several available software packages (reviewed in [6–8]). Artifacts due to insufficient preprocessing of the data have, in the worst case, led to erroneous biological conclusions in early SELDI studies [9–11]. This fact inspired several important comparison studies of SELDI preprocessing algorithms [6–8, 12]. We now briefly summarize a few of the major contributions. For a more detailed overview, see the introduction of .
Coombes et al introduced the use of wavelets for denoising SELDI spectra , providing a more adaptive approach to denoise compared to moving average filters (e.g., as in ). Meanwhile, Morris et al introduced the notion of a mean spectrum, which represents average protein activity of a group of spectra. Under non-restrictive assumptions, the mean spectrum has less noise and allows one to circumvent complicated peak matching algorithms that consolidate peak predictions among individual spectra into a consensus prediction. Malyarenko et al introduced a novel baseline removal algorithm based on a proposed charge accumulation model of the saturation phenomenon of the detector . This was one of the first algorithms that was designed from the "bottom-up", starting with physical considerations of SELDI. Later, deconvolution filters were shown to be a possible approach for improving mass resolution of SELDI [16–18].
Sköld et al analyzed single-shot spectra , the basic components of a final SELDI spectrum obtained by summing the results of many laser shots. They suggested that the observed counts in the single shot spectra may be proportional to a Poisson random variable, proposing a heteroscedastic model for the data. Meuleman et al also make use of single-shot spectra (sub-spectra) to derive a preprocessing algorithm based on analyzing these components separately .
In an attempt to improve on the bottom-up approach to preprocessing, we analyze the statistics of the SELDI signal over a wide range of intensity values. Based on data presented herein, we propose a natural exponential family model with quadratic variance function for the statistics of the detector response for SELDI experiments. We believe this model is a plausible explanation for acquisition of single-shot spectra, summing of single-shot spectra into a final spectrum, and extracting protein estimates from a mean spectrum under a unified framework. Under this framework, we introduce a new preprocessing approach, adaptive to changing noise characteristics per spectrum and per experiment, and show favorable peak prediction performance.
Buffer-only intensity measurements
Electronic measurements exhibit natural random fluctuations . In many cases, these fluctuations are independent of the signal and are modeled as additive white Gaussian noise. In order to understand the nature of the noise fluctuations inherent to SELDI, we study the response of the detector under controlled experiments applying different buffers instead of protein samples under varying laser intensities (as in ). This eliminates the complexity introduced by adding serum to the chips while facilitating measurements of ion counts over a wide range of intensity values. In principle, this gives us a set of n repeated experiments from which we can study the statistics of the detector response compounded with noise and interference inherent to SELDI. In this fashion, we have generated two separate buffer + matrix datasets, denoted BUFFER1 and BUFFER2, which represent data generated on the same SELDI PBS IIc machine by different scientists and different machine parameters. BUFFER1/BUFFER2 contain 183/114 spectra, respectively.
Figure 1 shows us the behavior of the typical buffer + baseline signal component seen in all SELDI raw spectra. Indeed, we see that changing different machine settings leads to different response properties. For BUFFER2, the median spectral response is large in the range shown, and the distribution of responses is symmetric about the median, whereas the distribution of detector response values for BUFFER1 are heavily skewed, and thus certainly not normally distributed.
Intensity fluctuation/variance increases monotonically with the mean.
The variance of the detector response is a quadratic function of the mean, to a very good approximation
The detector response curves for BUFFER1 and BUFFER2 are quite different, and thus are dependent on the machine settings.
with μ being the mean of X, V(μ) the variance, and υ 0 , υ1, υ2 constants, some of which may be zero.
From these observations, summarized in Figures 1 and 2, it seems unlikely that an algorithm optimized for BUFFER1 would work well on BUFFER2 and vice versa. Further, neither a homoscedastic approach (e.g. - standard wavelet shrinkage ) or a simple heteroscedastic approach (e.g. - Poisson regression formulation ) to preprocessing the data is likely to be sufficient.
Data for evaluating preprocessing algorithms
Exact protein content is known (and thus expectation of where "true" peaks will appear)
Analyzed sample is complex containing many proteins/peaks
Noise and baseline characteristics should be as close to those of real SELDI data as possible.
If one uses simulated data [6, 7, 25], complete control can be attained over requirements 1) and 2) at the expense of having noise/baseline characteristics that are overly ideal. If one uses purely real data, the noise, baseline, and artifacts that arise in actual experiments are present. However, this usually accompanies the trade-off of either not knowing the exact protein content (e.g.- complex serum data) or an overly simplified scenario (e.g. - spike-in data).
Further details on the hybrid spectra can be found in the Methods section and in Additional file 1. The collection of hybrid spectra under different operating conditions results in test sets, denoted HYBRID1 and HYBRID2, with each test set containing thirty datasets of fifty hybrid spectra each. The mean performance of a preprocessing algorithm on HYBRID1 and HYBRID2 can be interpreted as the expected performance of the preprocessing approach in each separate operating condition in a repeated experiment or sampling from a homogeneous population (e.g. - cancer group or control group).
New preprocessing algorithms for SELDI
We have developed a set of MATLAB® scripts for preprocessing SELDI spectra named LibSELDI. For information on how to obtain LibSELDI and the associated scripts used to produce the figures in this paper, contact the authors. We compare our preprocessing package to the MassSpecWavelet package from the Bioconductor project . MassSpecWavelet has been established as one of the best approaches in terms of peak finding in recent comparison studies [6, 7], and has been downloaded > 6000 times in the past two years as of March 2010 http://bioconductor.org/packages/stats/bioc/MassSpecWavelet.html. Both packages have the advantage of having only one main user-adjusted parameter.
Area under the operating characteristic comparison
At FDR values greater than 30%, MassSpecWavelet outperforms LibSELDI. However, this is at the expense of generally more promiscuous predictions, since MassSpecWavelet generates 586 potential protein predictions compared to 250 for LibSELDI.
If a random variable X ∈ NEF-QVF, it is completely specified by its variance function V(μ)
If X ∈ NEF-QVF, a, b constants then aX + b is also NEF-QVF
Additivity: If X 1; X 2 ∈ NEF-QVF, then X 1 + X 2 is NEF-QVF
Affine combinations of normal, Poisson, gamma, binomial, negative binomial, and generalized hyperbolic secant distributed random variables generate all possible distributions in the NEF-QVF family.
Constant Term: This is possibly due to thermal noise (additive Gaussian noise) which is common to all electronic measurement devices 
Linear Term: The ability to detect an ion in a multiple stage electron multiplier, a common type of detector in MALDI-like instruments, is described by compound Poisson statistics .
The existence of a plausible physical explanation for the quadratic variance term remains an open question. However its effect is measured in both BUFFER1 and BUFFER2 and cannot be neglected. While the QVF model explains the data well in the mass focused region between 3 and 30 kDa, it is likely to break down at lower masses around 2-2.5 kDA where the baseline reaches a maximum. In this region the detector often saturates, introducing a non-linearity into the data that we have not accounted for.
The success of our univariate model for SELDI may indicate that we have selected the most important feature to consider in the preprocessing of the data: namely, the fluctuations in the response of the ion detector subject to different inputs. The analysis of expression values of preprocessed data, on the other hand, requires multivariate methods as there are significant statistical dependencies between the peak heights corresponding to proteins that may be interacting. While these correlations are important in the analysis performed after the data is preprocessed, our results indicated it may be safe to ignore them during the preprocessing. While we have shown LibSELDI to be accurate for estimating peak m/z values, we have not assessed the usefulness approach for estimating peak intensities in this work. The utility of LibSELDI for accurately estimating peak intensities remains an open question and subject of future work.
It is entirely possible that the quadratic variance model could be applicable to other similar technologies such as MALDI and newer SELDI mass spectrometers. This, however, has not been confirmed.
Having buffer only spectra allows one to estimate the parameters of the detector response curve. Knowledge of the detector response curve enables us to apply the modified Antoniadis-Sapatinas denoising scheme described in the methods. Using this approach in our LibSELDI package yields excellent peak detection performance. We have proved this concept on HYBRID1 and HYBRID2 by estimating the QVF parameters of (1) using the buffer-only spectra that were randomly selected from BUFFER1 and BUFFER2 respectively. This implies that spots on SELDI chips should be reserved for buffer-only spectra. Thus, the trade-off for using our approach is increased cost in terms of the number of chips one must use. The modified Antoniadis-Sapatinas denoising is computationally intensive as well, taking approximately seven minutes per spectrum on a high-end workstation.
We argue that some of the cost is recovered by the potential for adaptive and accurate preprocessing, but not all. It may be possible to use QC and/or calibration samples to estimate the QVF as well rather than buffer-only spots. However, this would add in some additional variation due to the nature of the medium (serum, plasma, etc).
While LibSELDI outperforms MassSpecWavelet on the HYBRID1 and HYBRID2 test sets, the applicability of this comparison and of these results to purely real data remains an open question. There is some basic biological variability modeled in our test sets (see description in supplement of ). However, data from complex biological samples such as serum or plasma likely contains more biological variation and artifacts than we have modeled in HYBRID1 and HYBRID2. The investigation of how biological variation affects the model in QC samples is a work in progress.
We have shown that the variance of the intensity of a SELDI spectrum is quadratic in the mean signal strength. We further make the flexible assumption that the underlying distribution of the intensities is from a natural exponential family. From this point of view, we use a modified Antoniadis-Sapatinas wavelet shrinkage approach for denoising SELDI spectra. With this method at the core of our LibSELDI program for preprocessing SELDI data, we demonstrate excellent sensitivity at low false discovery rates. For applications that can tolerate higher false discovery rates, the MassSpecWavelet algorithm performs better in this region.
Our work has implications in the design of SELDI experiments. Namely, the modified Antoniadis-Sapatinas denoising technique performs well but requires an estimate of the quadratic variance function (QVF) describing the SELDI detector. This, in turn, is affected by machine settings. We have used buffer-only spectra to estimate the QVF. Thus, buffer-only spots could be interlaced on chips. We are investigating less expensive ways to estimate the QVF in future work.
Protocol for generating buffer-only spectra
Buffer-only spectra were generated by interspersing buffer only samples with protein samples from subjects (e.g. serum samples) and with pooled subject samples (for quality control) on the same chip. The buffer-only samples were spotted with wash buffer that was either PBS (phosphate buffered saline with various concentrations of phosphate and NaCl) based or acetonitrile + TFA (triflouroacetic acid) based, as manufacturer recommended per chip type. These buffer only samples were processed with the same washing steps as the subject samples, as described in , and then SPA matrix was applied to all spots.
The samples were analyzed with the Protein Biological System IIc™ SELDI mass spectrometer (Ciphergen Biosystems, Freemont, CA). The machine settings (e.g. laser intensity, detector sensitivity) and precise washing steps varied from buffer only spot to buffer only spot, and were generally different between BUFFER1 and BUFFER2. Note especially that laser intensities were generally higher for BUFFER2 than for BUFFER1. A detailed list of machine settings is given in the Additional file 1.
Calculating performance statistics for comparison of MassSpecWavelet and LibSELDI requires a large number of spectra emulating an experiment that was repeated many times. To generate the HYBRID1 dataset, we combine each clean spectrum with one buffer+matrix spectrum from BUFFER1, and similarly we form HYBRID2 from BUFFER2 by combining those spectra with the same clean spectra.
A basic model of repetitive experiments for SELDI is available with SimSpec 2.1 that takes into account fluctuations in protein concentrations, m/z values, and prevalence in the data. Using the SimSpec 2.1 model developed at the MD Anderson Cancer Center [25, 26], we generate 30 datasets containing 50 clean (noise and matrix-free) spectra each. Each dataset consists of 150 virtual proteins and each spectrum within the given dataset contains a proper subset of these proteins with fluctuating parameters according to the model described in  and its supplement. The goal for the preprocessing programs in our performance evaluation is to reconstruct the master list of 150 virtual proteins characterizing the dataset. Repeated across all 30 datasets, we can calculate useful performance statistics. The properties of the 150 virtual proteins themselves are drawn from a prior distribution that was estimated from real data. See , or alternatively, the description in the supplement of .
We use sampling to overcome the limitation of having much fewer spectra in BUFFER1 and BUFFER2 than we have clean spectra in preparation for testing the algorithms. In principle the best way to construct the hybrid test sets would be to have one unique spectrum in BUFFER1 (likewise BUFFER2) for each spectrum in our clean protein-only set. However, this would require 1500 buffer+matrix runs to be performed for both BUFFER1 and BUFFER2, an impractical amount of blank chips to run. Sampling from BUFFER1 (BUFFER2) provides a cost effective way to introduce variation in the noise/matrix characteristics between the datasets in HYBRID1 (HYBRID2).
Preprocessing the spectra
First we consider a model for a single SELDI spectrum, X(t). We observe X(t), a random process, on a discrete time grid t1,..., t m , where X(t) represents the intensity of the raw SELDI spectrum observed at time (equivalently mass) point t. For all t, we assume that X(t) is distributed according to a natural exponential family (NEF) with quadratic variance function (QVF) equal to V(μ(t)) as in Eq. (1). The variance function V(μ) completely characterizes the NEF-QVF family. The goal of preprocessing in SELDI is to estimate μ(t), the expectation of X(t), which is the signal corresponding to ions that hit the detector. With a good estimate of μ(t), extracting peaks and estimating protein m/z values in a dataset is relatively straightforward.
As a side note we point out that a SELDI spectrum is actually a sum of single shot spectra. However, the additivity property of the NEF-QVF family guarantees the sum is NEF-QVF provided that the single-shot spectra are NEF-QVF, agreeing with our detector response model and experimental observations.
Multiple spectra considerations
Thus, the mean spectrum concept is valuable under the assumptions of the NEF-QVF model as well.
Modified Antoniadis-Sapatinas denoising
Where V(x • ) is the vector constructed by applying the QVF from (1) to each term of x • . (W · W ) is the matrix whose i, j element is the square of the i, j element of W. The parameters υ0, υ1, υ2 in (1) are measured from the buffer-only spectra, as described in the Results and Discussion section.
Peak detection/baseline removal
where, denotes a Gaussian kernel function centered at t j with standard deviation σ j and zero outside the interval [t j - α, t j + α].
Typically, s(t) is very sparse in the sense that it is mostly zero over the domain of the observed signal. Therefore, the local minima of our estimated baseline + noise signal are points we may assume touch the baseline. From this point of view, once we have detected all the local minima in , the baseline curve estimation problem reduces to an interpolation problem amongst these points. We have found through experimentation that piecewise cubic Hermite interpolating polynomials  are excellent interpolation functions.
The minima and maxima in are found in one pass using the extrema function downloadable from MATLAB® central file exchange. The maxima are the peaks in the mean spectrum potentially indicating proteins represented in our sample population while the minima correspond to samples from the baseline signal.
Each detected peak is quantified using peak area and a threshold is chosen based on the peak area measurement to generate the final prediction set.
Where TP (the number of true positives) is the number of the 150 virtual protein m/z values having at least one predicted m/z value within 0.3% relative error. The FP is defined as the number of predicted m/z values not within 0.3% of any of the 150 virtual protein m/z values for this dataset. Similarly, FN is the number of the 150 virtual protein values without any predicted m/z value within 0.3% relative error.
For each dataset, a curve is fit to the operating points. Each operating curve is averaged to produce a mean operating characteristic, as shown in Figure 4. From this curve, the calculation of the area-under-the curve is straightforward. For more details, see sections 2.2 and 2.2.1 of .
The authors would like to thank Dr. Gitika Panicker and Dr. Dominique Rollin of the Centers for Disease Control and Prevention, Atlanta, GA, USA for conception, generation, and acquisition of the buffer+matrix data used for BUFFER1 and BUFFER2, respectively. The authors would also like to thank Dr. Sally Lin and Dr. Toni Whistler (also at the CDC) for useful discussions that led to improvements in the manuscript. Lastly, we would to thank Dr. Elizabeth Unger and Dr. William C. Reeves of the CDC and Dr. G. Tong Zhou of Georgia Tech for support during the research and writing of the manuscript.
This work was supported in part by NCI's Early Detection Research Network through IAA Y1-CN-5005-01 and Y1-CN-0101-01.
This research was supported in part by an appointment to the Research Participation Program at the Centers for Disease Control and Prevention, National Center for Emerging and Zoonotic Infectious Diseases, Division of High-Consequence Pathogens and Pathology, Chronic Viral Diseases Branch, administered by the Oak Ridge Institute for Science and Education through an inter-agency agreement between the U.S. Department of Energy and the CDC.
The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the funding agencies. The authors declare no conflict of interest.
- Hüttenhain R, Malmström J, Picotti P, Aebersold R: Perspectives of targeted mass spectrometry for protein biomarker verification. Curr Opin Chem Biol 2009, 13(5–6):518–525. 10.1016/j.cbpa.2009.09.014View ArticlePubMedPubMed CentralGoogle Scholar
- Domon B, Aebersold R: Mass spectrometry and protein analysis. Science 2006, 312(5771):212–217. 10.1126/science.1124619View ArticlePubMedGoogle Scholar
- Corthals GL, Wasinger VC, Hochstrasser DF, Sanchez JC: The dynamic range of protein expression: a challenge for proteomic research. Electrophoresis 2000, 21(6):1104–1115. 10.1002/(SICI)1522-2683(20000401)21:6<1104::AID-ELPS1104>3.0.CO;2-CView ArticlePubMedGoogle Scholar
- Hutchens TW, Yip T: New Desorption Strategies for the Mass Spectrometric Analysis of Macromolecules. Rapid Commun Mass Spectrom 1993, 7(7):576–580. 10.1002/rcm.1290070703View ArticleGoogle Scholar
- Bock MD, de Seny D, Meuwis MA, Chapelle JP, Louis E, Malaise M, Merville MP, Fillet M: Challenges for biomarker discovery in body fluids using SELDI-TOF-MS. J Biomed Biotechnol 2010, 2010: 906082.PubMedPubMed CentralGoogle Scholar
- Emanuele VA, Gurbaxani BM: Benchmarking currently available SELDI-TOF MS preprocessing techniques. Proteomics 2009, 9(7):1754–1762. 10.1002/pmic.200701171View ArticlePubMedGoogle Scholar
- Cruz-Marcelo A, Guerra R, Vannucci M, Li Y, Lau CC, Man TK: Comparison of algorithms for pre-processing of SELDI-TOF mass spectrometry data. Bioinformatics 2008, 24(19):2129–2136. 10.1093/bioinformatics/btn398View ArticlePubMedPubMed CentralGoogle Scholar
- Wegdam W, Moerland PD, Buist MR, van Themaat EVL, Bleijlevens B, Hoefsloot HCJ, de Koster CG, Aerts JMFG: Classification-based comparison of pre-processing methods for interpretation of mass spectrometry generated clinical datasets. Proteome Sci 2009, 7: 19. 10.1186/1477-5956-7-19View ArticlePubMedPubMed CentralGoogle Scholar
- Baggerly KA, Morris JS, Coombes KR: Reproducibility of SELDI-TOF protein patterns in serum: comparing datasets from different experiments. Bioinformatics 2004, 20(5):777–785. 10.1093/bioinformatics/btg484View ArticlePubMedGoogle Scholar
- Sorace JM, Zhan M: A data review and re-assessment of ovarian cancer serum proteomic profiling. BMC Bioinformatics 2003, 4: 24. 10.1186/1471-2105-4-24View ArticlePubMedPubMed CentralGoogle Scholar
- Diamandis EP: Analysis of serum proteomic patterns for early cancer diagnosis: drawing attention to potential problems. J Natl Cancer Inst 2004, 96(5):353–356. 10.1093/jnci/djh056View ArticlePubMedGoogle Scholar
- Meuleman W, Engwegen JY, Gast MCW, Beijnen JH, Reinders MJ, Wessels LF: Comparison of normalisation methods for surface-enhanced laser desorption and ionisation (SELDI) time-of-flight (TOF) mass spectrometry data. BMC Bioinformatics 2008, 9: 88. 10.1186/1471-2105-9-88View ArticlePubMedPubMed CentralGoogle Scholar
- Coombes KR, Tsavachidis S, Morris JS, Baggerly KA, Hung MC, Kuerer HM: Improved peak detection and quantification of mass spectrometry data acquired from surface-enhanced laser desorption and ionization by denoising spectra with the undecimated discrete wavelet transform. Proteomics 2005, 5(16):4107–4117. 10.1002/pmic.200401261View ArticlePubMedGoogle Scholar
- Fung ET, Enderwick C: ProteinChip clinical proteomics: computational challenges and solutions. BioTechniques 2002, Suppl: 34–38. 40–41 40-41PubMedGoogle Scholar
- Malyarenko DI, Cooke WE, Adam BL, Malik G, Chen H, Tracy ER, Trosset MW, Sasinowski M, Semmes OJ, Manos DM: Enhancement of sensitivity and resolution of surface-enhanced laser desorption/ionization time-of-flight mass spectrometric records for serum peptides using time-series analysis techniques. Clin Chem 2005, 51: 65–74. 10.1373/clinchem.2004.037283View ArticlePubMedPubMed CentralGoogle Scholar
- Malyarenko D, Cooke W, Tracy E, Trosset M, Semmes O, Sasinowski M, Manos D: Deconvolution filters to enhance resolution of dense time-of-flight survey spectra in the time-lag optimization range. Rapid Commun Mass Spectrom 2006, 20(11):1661–9. [Deconvolution filters; dense time-of-flight survey spectral resolution;time-lag optimization range;time-domain filters; time-of-flight mass spectrometry signals;signal-to-noise ratio; filtering;smoothing;time series method;matrix-assisted laser desorption/ionization;biomolecules;nonlinear filters;] [Deconvolution filters; dense time-of-flight survey spectral resolution;time-lag optimization range;time-domain filters; time-of-flight mass spectrometry signals;signal-to-noise ratio; filtering;smoothing;time series method;matrix-assisted laser desorption/ionization;biomolecules;nonlinear filters;] 10.1002/rcm.2487View ArticlePubMedPubMed CentralGoogle Scholar
- Malyarenko DI, Cooke WE, Tracy ER, Drake RR, Shin S, Semmes OJ, Sasinowski M, Manos DM: Resampling and deconvolution of linear time-of-flight records for enhanced protein profiling. Rapid Commun Mass Spectrom 2006, 20(11):1670–1678. 10.1002/rcm.2496View ArticlePubMedGoogle Scholar
- Gatlin-Bunai CL, Cazares LH, Cooke WE, Semmes OJ, Malyarenko DI: Optimization of MALDI-TOF MS detection for enhanced sensitivity of affinity-captured proteins spanning a 100 kDa mass range. J Proteome Res 2007, 6(11):4517–4524. 10.1021/pr0703526View ArticlePubMedPubMed CentralGoogle Scholar
- Sköld M, Rydén T, Samuelsson V, Bratt C, Ekblad L, Olsson H, Baldetorp B: Regression analysis and modelling of data acquisition for SELDI-TOF mass spectrometry. Bioinformatics 2007, 23(11):1401–1409. 10.1093/bioinformatics/btm104View ArticlePubMedGoogle Scholar
- Meuleman W, Engwegen JYMN, Gast MCW, Wessels LFA, Reinders MJT: Analysis of mass spectrometry data using sub-spectra. BMC Bioinformatics 2009, 10(Suppl 1):S51. 10.1186/1471-2105-10-S1-S51View ArticlePubMedPubMed CentralGoogle Scholar
- van der Ziel A: Noise in Measurements. Wiley-Interscience; 1976.Google Scholar
- Rollin D, Whistler T, Vernon SD: Laboratory methods to improve SELDI peak detection and quantitation. Proteome Sci 2007, 5: 9. 10.1186/1477-5956-5-9View ArticlePubMedPubMed CentralGoogle Scholar
- Donoho DL, Johnstone IM: Ideal spatial adaptation by wavelet shrinkage. Biometrika 1994, 81: 425–455. 10.1093/biomet/81.3.425View ArticleGoogle Scholar
- Donoho DL: Nonlinear Wavelet Methods for Recovery of Signals, Densities, and Spectra from Indirect and Noisy Data. In In Proceedings of Symposia in Applied Mathematics. American Mathematical Society; 1993:173–205.Google Scholar
- Morris JS, Coombes KR, Koomen J, Baggerly KA, Kobayashi R: Feature extraction and quantification for mass spectrometry in biomedical applications using the mean spectrum. Bioinformatics 2005, 21(9):1764–1775. 10.1093/bioinformatics/bti254View ArticlePubMedGoogle Scholar
- Coombes KR, Koomen JM, Baggerly KA, Morris JS, Kobayashi R: Understanding the characteristics of mass spectrometry data through the use of simulation. Cancer Informatics 2005, 1: 41–52.PubMedPubMed CentralGoogle Scholar
- Du P, Kibbe WA, Lin SM: Improved peak detection in mass spectrum by incorporating continuous wavelet transform-based pattern matching. Bioinformatics 2006, 22(17):2059–2065. 10.1093/bioinformatics/btl355View ArticlePubMedGoogle Scholar
- Morris CN: Natural exponential families with quadratic variance functions. The Annals of Statistics 1982, 10: 65–80. 10.1214/aos/1176345690View ArticleGoogle Scholar
- Dietz LA: Basic Properties of Electron Multiplier Ion Detection and Pulse Counting Methods in Mass Spectrometry. Rev Sci Instrum 1965, 36(12):1763–1770. 10.1063/1.1719460View ArticleGoogle Scholar
- Antoniadis A, Sapatinas T: Wavelet shrinkage for natural exponential families with quadratic variance functions. Biometrika 2001, 88(3):805–820. 10.1093/biomet/88.3.805View ArticleGoogle Scholar
- Daubechies I: Ten lectures on wavelets. Philadelphia, PA, USA: Society for Industrial and Applied Mathematics; 1992.View ArticleGoogle Scholar
- Fritsch FN, Carlson RE: Monotone Piecewise Cubic Interpolation. SIAM j Numerical Analysis 1980, 17: 238–246. 10.1137/0717021View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.