A Bayesian network approach to feature selection in mass spectrometry data.

BACKGROUND: Time-of-flight mass spectrometry (TOF-MS) has the potential to provide non-invasive, high-throughput screening for cancers and other serious diseases via detection of protein biomarkers in blood or other accessible biologic samples. Unfortunately, this potential has largely been unrealized to date due to the high variability of measurements, uncertainties in the distribution of proteins in a given population, and the difficulty of extracting repeatable diagnostic markers using current statistical tools. With studies consisting of perhaps only dozens of samples, and possibly hundreds of variables, overfitting is a serious complication. To overcome these difficulties, we have developed a Bayesian inductive method which uses model-independent methods of discovering relationships between spectral features. This method appears to efficiently discover network models which not only identify connections between the disease and key features, but also organizes relationships between features--and furthermore creates a stable classifier that categorizes new data at predicted error rates. RESULTS: The method was applied to artificial data with known feature relationships and typical TOF-MS variability introduced, and was able to recover those relationships nearly perfectly. It was also applied to blood sera data from a 2004 leukemia study, and showed high stability of selected features under cross-validation. Verification of results using withheld data showed excellent predictive power. The method showed improvement over traditional techniques, and naturally incorporated measurement uncertainties. The relationships discovered between features allowed preliminary identification of a protein biomarker which was consistent with other cancer studies and later verified experimentally. CONCLUSIONS: This method appears to avoid overfitting in biologic data and produce stable feature sets in a network model. The network structure provides additional information about the relationships among features that is useful to guide further biochemical analysis. In addition, when used to classify new data, these feature sets are far more consistent than those produced by many traditional techniques.

Precision enhancement of MALDI-TOF MS using high resolution peak detection and label-free alignment.

We have developed an automated procedure for aligning peaks in multiple TOF spectra that eliminates common timing errors and small variations in spectrometer output. Our method incorporates high-resolution peak detection, re-binning, and robust linear data fitting in the time domain. This procedure aligns label-free (uncalibrated) peaks to minimize the variation in each peak's location from one spectrum to the next, while maintaining a high number of degrees of freedom. We apply our method to replicate pooled-serum spectra from multiple laboratories and increase peak precision (t/sigma(t)) to values limited only by small random errors (with sigma(t) less than one time count in 89 out of 91 instances, 13 peaks in seven datasets). The resulting high precision allowed for an order of magnitude improvement in peak m/z reproducibility. We show that the CV for m/z is 0.01% (100 ppm) for 12 out of the 13 peaks that were observed in all datasets between 2995 and 9297 Da.

Improved signal processing and normalization for biomarker protein detection in broad-mass-range TOF mass spectra from clinical samples.

PURPOSE: To demonstrate robust detection of biomarkers in broad-mass-range TOF-MS data. EXPERIMENTAL DESIGN: Spectra were obtained for two serum protein profiling studies: (i) 2-200 kDa for 132 patients, 67 healthy and 65 diagnosed as having adult T-cell leukemia and (ii) 2-100 kDa for 140 patients, 70 pairs, each with matched prostate-specific antigen (PSA) levels and biopsy-confirmed diagnoses of one benign and one prostate cancer. Signal processing was performed on raw spectra and peak data were normalized using four methods. Feature selection was performed using Bayesian Network Analysis and a classifier was tested on withheld data. Identification of candidate biomarkers was pursued. RESULTS: Integrated peak intensities were resolved over full spectra. Normalization using local noise values was superior to global methods in reducing peak correlations, reducing replicate variability and improving feature selection stability. For the leukemia data set, potential disease biomarkers were detected and were found to be predictive for withheld data. Preliminary assignments of protein IDs were consistent with published results and LC-MS/MS identification. No prostate-specific-antigen-independent biomarkers were detected in the prostate cancer data set. CONCLUSIONS AND CLINICAL RELEVANCE: Signal processing, local signal-to-noise (SNR) normalization and Bayesian Network Analysis feature selection facilitate robust detection and identification of biomarker proteins in broad-mass-range clinical TOF-MS data.

Deconvolution filters to enhance resolution of dense time-of-flight survey spectra in the time-lag optimization range.

By applying time-domain filters to time-of-flight (TOF) mass spectrometry signals, we have simultaneously smoothed and narrowed spectra resulting in improved resolution and increased signal-to-noise ratios. This filtering procedure has an advantage over detailed curve fitting of spectra in the case of large dense spectra, when neither the location nor the number of mass peaks is known a priori. This time series method is directly applicable in the time lag optimization range, where point density per peak is constant. We present a systematic methodology to optimize the filters according to any desired figure of merit, illustrating the procedure by optimizing the signal-to-noise per unit bandwidth of matrix-assisted laser desorption/ionization (MALDI) data. We also introduce a nonlinear filter that reduces the spurious structure that often accompanies deconvolution filters. The net result of the application of these filters is that we can identify new structures in dense MALDI-TOF data, clearly showing small adducts to heavy biomolecules.

Resampling and deconvolution of linear time-of-flight records for enhanced protein profiling.

We have developed a peak deconvolution strategy that is applicable to the full mass range of a time-of-flight (TOF) spectrum. This strategy involves resampling a spectrum to create a time series that has equal peak widths (in time) across the entire spectrum, and then using the deconvolution filters we have previously described. We use this technique to deconvolve the protein mass spectra for blood serum and cell lysates acquired on three separate TOF instruments. Following deconvolution, we resolve spectral structures consistent with expected events such as multiply charged ions, matrix adducts and post-translational protein modifications. The deconvolution procedure produces a 40% improvement in the resolution and enhanced experimental sensitivity over the full length of the linear TOF record, up to m/z 150 000. This approach is particularly appropriate for automated data analysis and peak detection in dense TOF spectra.

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.

BACKGROUND: Measurement of peptide/protein concentrations in biological samples for biomarker discovery commonly uses high-sensitivity mass spectrometers with a surface-processing procedure to concentrate the important peptides. These time-of-flight (TOF) instruments typically have low mass resolution and considerable electronic noise associated with their detectors. The net result is unnecessary overlapping of peaks, apparent mass jitter, and difficulty in distinguishing mass peaks from background noise. Many of these effects can be reduced by processing the signal using standard time-series background subtraction, calibration, and filtering techniques. METHODS: Surface-enhanced laser desorption/ionization (SELDI) spectra were acquired on a PBS II instrument from blank, hydrophobic, and IMAC-Cu ProteinChip arrays (Ciphergen Biosystems, Inc.) incubated with calibration peptide mixtures or pooled serum. TOF data were recorded after single and multiple laser shots at different positions. Correlative analysis was used for time-series calibration. Target filters were used to suppress noise and enhance resolution after baseline removal and noise rescaling. RESULTS: The developed algorithms compensated for the electronic noise attributable to detector overload, removed the baseline caused by charge accumulation, detected and corrected mass peak jitter, enhanced signal amplitude at higher masses, and improved the resolution by using a deconvolution filter. CONCLUSIONS: These time-series techniques, when applied to SELDI-TOF data before any peak identification procedure, can improve the data to make the peak identification process simpler and more robust. These improvements may be applicable to most TOF instrumentation that uses analog (rather than counting) detectors.

Ray helicity: geometric invariant for multidimensional resonant wave conversion.

For a multicomponent wave field propagating into a multidimensional conversion region, the rays are shown to be helical, in general. For a ray-based quantity to have a fundamental physical meaning, it must be invariant under the following two groups of transformations, which are used to construct solutions: congruence transformations (which involve linear combinations of components of the multicomponent wave field) and canonical transformations (which act on the ray phase space). It is shown that for conversion between two waves there is a new invariant not previously discussed: the intrinsic helicity of the ray.