Robust prediction of the MASCOT score for an improved quality assessment in mass spectrometric proteomics

Protein identification by tandem mass spectrometry is based on the reliable processing of the acquired data. Unfortunately, the generation of a large number of poor quality spectra is commonly observed in LC-MS/MS, and the processing of these mostly noninformative spectra with its associated costs should be avoided. We present a continuous quality score that can be computed very quickly and that can be considered an approximation of the MASCOT score in case of a correct identification. This score can be used to reject low quality spectra prior to database identification, or to draw attention to those spectra that exhibit a (supposedly) high information content, but could not be identified. The proposed quality score can be calibrated automatically on site without the need for a manually generated training set. When this score is turned into a classifier and when features are used that are independent of the instrument, the proposed approach performs equally to previously published classifiers and feature sets and also gives insights into the behavior of the MASCOT score.     

The standard protein mix database: a diverse data set to assist in the production of improved Peptide and protein identification software tools

Tandem mass spectrometry (MS/MS) is frequently used in the identification of peptides and proteins. Typical proteomic experiments rely on algorithms such as SEQUEST and MASCOT to compare thousands of tandem mass spectra against the theoretical fragment ion spectra of peptides in a database. The probabilities that these spectrum-to-sequence assignments are correct can be determined by statistical software such as PeptideProphet or through estimations based on reverse or decoy databases. However, many of the software applications that assign probabilities for MS/MS spectra to sequence matches were developed using training data sets from 3D ion-trap mass spectrometers. Given the variety of types of mass spectrometers that have become commercially available over the last 5 years, we sought to generate a data set of reference data covering multiple instrumentation platforms to facilitate both the refinement of existing computational approaches and the development of novel software tools. We analyzed the proteolytic peptides in a mixture of tryptic digests of 18 proteins, named the "ISB standard protein mix", using 8 different mass spectrometers. These include linear and 3D ion traps, two quadrupole time-of-flight platforms (qq-TOF), and two MALDI-TOF-TOF platforms. The resulting data set, which has been named the Standard Protein Mix Database, consists of over 1.1 million spectra in 150+ replicate runs on the mass spectrometers. The data were inspected for quality of separation and searched using SEQUEST. All data, including the native raw instrument and mzXML formats and the PeptideProphet validated peptide assignments, are available at http://regis-web.systemsbiology.net/PublicDatasets/.     

HMMatch: peptide identification by spectral matching of tandem mass spectra using hidden Markov models.

Peptide identification by tandem mass spectrometry is the dominant proteomics workflow for protein characterization in complex samples. The peptide fragmentation spectra generated by these workflows exhibit characteristic fragmentation patterns that can be used to identify the peptide. In other fields, where the compounds of interest do not have the convenient linear structure of peptides, fragmentation spectra are identified by comparing new spectra with libraries of identified spectra, an approach called spectral matching. In contrast to sequence-based tandem mass spectrometry search engines used for peptides, spectral matching can make use of the intensities of fragment peaks in library spectra to assess the quality of a match. We evaluate a hidden Markov model approach (HMMatch) to spectral matching, in which many examples of a peptide's fragmentation spectrum are summarized in a generative probabilistic model that captures the consensus and variation of each peak's intensity. We demonstrate that HMMatch has good specificity and superior sensitivity, compared to sequence database search engines such as X!Tandem. HMMatch achieves good results from relatively few training spectra, is fast to train, and can evaluate many spectra per second. A statistical significance model permits HMMatch scores to be compared with each other, and with other peptide identification tools, on a unified scale. HMMatch shows a similar degree of concordance with X!Tandem, Mascot, and NIST's MS Search, as they do with each other, suggesting that each tool can assign peptides to spectra that the others miss. Finally, we show that it is possible to extrapolate HMMatch models beyond a single peptide's training spectra to the spectra of related peptides, expanding the application of spectral matching techniques beyond the set of peptides previously observed.     

Linear discriminant analysis-based estimation of the false discovery rate for phosphopeptide identifications

The development of liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) has made it possible to characterize phosphopeptides in an increasingly large-scale and high-throughput fashion. However, extracting confident phosphopeptide identifications from the resulting large data sets in a similar high-throughput fashion remains difficult, as does rigorously estimating the false discovery rate (FDR) of a set of phosphopeptide identifications. This article describes a data analysis pipeline designed to address these issues. The first step is to reanalyze phosphopeptide identifications that contain ambiguous assignments for the incorporated phosphate(s) to determine the most likely arrangement of the phosphate(s). The next step is to employ an expectation maximization algorithm to estimate the joint distribution of the peptide scores. A linear discriminant analysis is then performed to determine how to optimally combine peptide scores (in this case, from SEQUEST) into a discriminant score that possesses the maximum discriminating power. Based on this discriminant score, the p- and q-values for each phosphopeptide identification are calculated, and the phosphopeptide identification FDR is then estimated. This data analysis approach was applied to data from a study of irradiated human skin fibroblasts to provide a robust estimate of FDR for phosphopeptides. The Phosphopeptide FDR Estimator software is freely available for download at http://ncrr.pnl.gov/software/.     

Adaptive discriminant function analysis and reranking of MS/MS database search results for improved peptide identification in shotgun proteomics

Robust statistical validation of peptide identifications obtained by tandem mass spectrometry and sequence database searching is an important task in shotgun proteomics. PeptideProphet is a commonly used computational tool that computes confidence measures for peptide identifications. In this paper, we investigate several limitations of the PeptideProphet modeling approach, including the use of fixed coefficients in computing the discriminant search score and selection of the top scoring peptide assignment per spectrum only. To address these limitations, we describe an adaptive method in which a new discriminant function is learned from the data in an iterative fashion. We extend the modeling framework to go beyond the top scoring peptide assignment per spectrum. We also investigate the effect of clustering the spectra according to their spectrum quality score followed by cluster-specific mixture modeling. The analysis is carried out using data acquired from a mixture of purified proteins on four different types of mass spectrometers, as well as using a complex human serum data set. A special emphasis is placed on the analysis of data generated on high mass accuracy instruments.     

A graph-theoretic approach for the separation of b and y ions in tandem mass spectra

MOTIVATION: Ion-type identification is a fundamental problem in computational proteomics. Methods for accurate identification of ion types provide the basis for many mass spectrometry data interpretation problems, including (a) de novo sequencing, (b) identification of post-translational modifications and mutations and (c) validation of database search results. RESULTS: Here, we present a novel graph-theoretic approach for solving the problem of separating b ions from y ions in a set of tandem mass spectra. We represent each spectral peak as a node and consider two types of edges: type-1 edge connecting two peaks probably of the same ion types and type-2 edge connecting two peaks probably of different ion types. The problem of ion-separation is formulated and solved as a graph partition problem, which is to partition the graph into three subgraphs, representing b, y and others ions, respectively, through maximizing the total weight of type-1 edges while minimizing the total weight of type-2 edges within each partitioned subgraph. We have developed a dynamic programming algorithm for rigorously solving this graph partition problem and implemented it as a computer program PRIME (PaRtition of Ion types in tandem Mass spEctra). The tests on a large amount of simulated mass spectra and 19 sets of high-quality experimental Fourier transform ion cyclotron resonance tandem mass spectra indicate that an accuracy level of approximately 90% for the separation of b and y ions was achieved. AVAILABILITY: The executable code of PRIME is available upon request. CONTACT: xyn@bmb.uga.edu.     

A comprehensive approach to the analysis of matrix-assisted laser desorption/ionization-time of flight proteomics spectra from serum samples

For our analysis of the data from the First Annual Proteomics Data Mining Conference, we attempted to discriminate between 24 disease spectra (group A) and 17 normal spectra (group B). First, we processed the raw spectra by (i) correcting for additive sinusoidal noise (periodic on the time scale) affecting most spectra, (ii) correcting for the overall baseline level, (iii) normalizing, (iv) recombining fractions, and (v) using variable-width windows for data reduction. Also, we identified a set of polymeric peaks (at multiples of 180.6 Da) that is present in several normal spectra (B1-B8). After data processing, we found the intensities at the following mass to charge (m/z) values to be useful discriminators: 3077, 12 886 and 74 263. Using these values, we were able to achieve an overall classification accuracy of 38/41 (92.6%). Perfect classification could be achieved by adding two additional peaks, at 2476 and 6955. We identified these values by applying a genetic algorithm to a filtered list of m/z values using Mahalanobis distance between the group means as a fitness function.     

Code developments to improve the efficiency of automated MS/MS spectra interpretation

We report the results of our work to facilitate protein identification using tandem mass spectra and protein sequence databases. We describe a parallel version of SEQUEST (SEQUEST-PVM) that is tolerant toward arithmetic exceptions. The changes we report effectively separate search processes on slave nodes from each other. Therefore, if one of the slave nodes drops out of the cluster due to an error, the rest of the cluster will carry the search process to the end. SEQUEST has been widely used for protein identifications. The modifications made to the code improve its stability and effectiveness in a high-throughput production environment. We evaluate the overhead associated with the parallelization of SEQUEST. A prior version of software to preprocess LC/MS/MS data attempted to differentiate the charge states of ions. Singly charged ions can be accurately identified, but the software was unable to reliably differentiate tandem mass spectra of +2 and +3 charge states. We have designed and implemented a computational approach to narrow charge states of precursor ions from nominal resolution ion-trap tandem mass spectra. The preprocessing code, 2to3, determines the charge state of the precursor ion using its mass-to-charge ratio (m/z) and fragment ions contained in the tandem mass spectrum. For each possible charge state the program calculates the expected fragment ions that account for precursor ion m/z values. If any one of the numbers is less than an empirically determined threshold value then the spectrum corresponding to that charge state is removed. If both numbers are higher than the threshold value then +2 and +3 copies of the spectrum are kept. We present the comparison of results from protein identification experiments with and without using 2 to 3. It is shown that by determining the charge state and eliminating poor quality spectra 2to3 decreases the number of spectral files to be searched without affecting the search results. The decrease reduces computer requirements and researcher efforts for analysis of the results.     

Probabilistic disease classification of expression-dependent proteomic data from mass spectrometry of human serum.

We have developed an algorithm called Q5 for probabilistic classification of healthy versus disease whole serum samples using mass spectrometry. The algorithm employs principal components analysis (PCA) followed by linear discriminant analysis (LDA) on whole spectrum surface-enhanced laser desorption/ionization time of flight (SELDI-TOF) mass spectrometry (MS) data and is demonstrated on four real datasets from complete, complex SELDI spectra of human blood serum. Q5 is a closed-form, exact solution to the problem of classification of complete mass spectra of a complex protein mixture. Q5 employs a probabilistic classification algorithm built upon a dimension-reduced linear discriminant analysis. Our solution is computationally efficient; it is noniterative and computes the optimal linear discriminant using closed-form equations. The optimal discriminant is computed and verified for datasets of complete, complex SELDI spectra of human blood serum. Replicate experiments of different training/testing splits of each dataset are employed to verify robustness of the algorithm. The probabilistic classification method achieves excellent performance. We achieve sensitivity, specificity, and positive predictive values above 97% on three ovarian cancer datasets and one prostate cancer dataset. The Q5 method outperforms previous full-spectrum complex sample spectral classification techniques and can provide clues as to the molecular identities of differentially expressed proteins and peptides.     

A Practical and Automated Approach to Large Area Forest Disturbance Mapping with Remote Sensing

In this paper, I describe a set of procedures that automate forest disturbance mapping using a pair of Landsat images. The approach is built on the traditional pair-wise change detection method, but is designed to extract training data without user interaction and uses a robust classification algorithm capable of handling incorrectly labeled training data. The steps in this procedure include: i) creating masks for water, non-forested areas, clouds, and cloud shadows; ii) identifying training pixels whose value is above or below a threshold defined by the number of standard deviations from the mean value of the histograms generated from local windows in the short-wave infrared (SWIR) difference image; iii) filtering the original training data through a number of classification algorithms using an n-fold cross validation to eliminate mislabeled training samples; and finally, iv) mapping forest disturbance using a supervised classification algorithm. When applied to 17 Landsat footprints across the U.S. at five-year intervals between 1985 and 2010, the proposed approach produced forest disturbance maps with 80 to 95% overall accuracy, comparable to those obtained from traditional approaches to forest change detection. The primary sources of mis-classification errors included inaccurate identification of forests (errors of commission), issues related to the land/water mask, and clouds and cloud shadows missed during image screening. The approach requires images from the peak growing season, at least for the deciduous forest sites, and cannot readily distinguish forest harvest from natural disturbances or other types of land cover change. The accuracy of detecting forest disturbance diminishes with the number of years between the images that make up the image pair. Nevertheless, the relatively high accuracies, little or no user input needed for processing, speed of map production, and simplicity of the approach make the new method especially practical for forest cover change analysis over very large regions.     

pParse: a method for accurate determination of monoisotopic peaks in high-resolution mass spectra

Determining the monoisotopic peak of a precursor is a first step in interpreting mass spectra, which is basic but non-trivial. The reason is that in the isolation window of a precursor, other peaks interfere with the determination of the monoisotopic peak, leading to wrong mass-to-charge ratio or charge state. Here we propose a method, named pParse, to export the most probable monoisotopic peaks for precursors, including co-eluted precursors. We use the relationship between the position of the highest peak and the mass of the first peak to detect candidate clusters. Then, we extract three features to sort the candidate clusters: (i) the sum of the intensity, (ii) the similarity of the experimental and the theoretical isotopic distribution, and (iii) the similarity of elution profiles. We showed that the recall of pParse, MaxQuant, and BioWorks was 98-98.8%, 0.5-17%, and 1.8-36.5% at the same precision, respectively. About 50% of tandem mass spectra are triggered by multiple precursors which are difficult to identify. Then we design a new scoring function to identify the co-eluted precursors. About 26% of all identified peptides were exclusively from co-eluted peptides. Therefore, accurately determining monoisotopic peaks, including co-eluted precursors, can greatly increase peptide identification rate.     

Optimal construction of theoretical spectra for MS/MS spectra identification

We derive the optimal number of peaks (defined as the minimum number that provides the required efficiency of spectra identification) in the theoretical spectra as a function of (i) the experimental accuracy, sigma, of the measured ratio m/z; (ii) experimental spectrum density; (iii) size of the database; (iv) number of peaks in the theoretical spectra; and (v) types of ions that the peaks represent. We show that if theoretical spectra are constructed including b and y ions alone, then for sigma = 0.5, which is typical for high-throughput data, peptide chains of eight amino acids or longer can be identified based on the positions of peaks alone, at a rate of false identification below 1%. To discriminate between shorter peptides, additional (e.g., intensity-inferred) information is necessary. We derive the dependence of the probability of false identification on the number of peaks in the theoretical spectra and on the types of ions that the peaks represent. Our results suggest that the class of mass spectrum identification problems, for which more elaborate development of fragmentation rules (such as intensity model) is required, can be reduced to the problems that involve homologous peptides.     

Improving feature detection and analysis of surface-enhanced laser desorption/ionization-time of flight mass spectra

Discovering valid biological information from surface-enhanced laser desorption/ionization-time of flight mass spectrometry (SELDI-TOF MS) depends on clear experimental design, meticulous sample handling, and sophisticated data processing. Most published literature deals with the biological aspects of these experiments, or with computer-learning algorithms to locate sets of classifying biomarkers. The process of locating and measuring proteins across spectra has received less attention. This process should be tunable between sensitivity and false-discovery, and should guarantee that features are biologically meaningful in that they represent chemical species that can be identified and investigated. Existing feature detection in SELDI-TOF MS is not optimal for acquiring biologically relevant data. Most methods have so many user-defined settings that reproducibility and comparability among studies suffer considerably. To address these issues, we have developed an approach, called simultaneous spectrum analysis (SSA), which (i) locates proteins across spectra, (ii) measures their abundance, (iii) subtracts baseline, (iv) excludes irreproducible measurements, and (v) computes normalization factors for comparing spectra. SSA uses only two key parameters for feature detection and one parameter each for quality thresholds on spectra and peaks. The effectiveness of SSA is demonstrated by identifying proteins differentially expressed in SELDI-TOF spectra from plasma of wild-type and knockout mice for plasma glutathione peroxidase. Comparing analyses by SSA and CiphergenExpress Data Manager 2.1 finds similar results for large signal peaks, but SSA improves the number and quality of differences betweens groups among lower signal peaks. SSA is also less likely to introduce systematic bias when normalizing spectra.     

Machine learning approaches to lung cancer prediction from mass spectra

We addressed the problem of discriminating between 24 diseased and 17 healthy specimens on the basis of protein mass spectra. To prepare the data, we performed mass to charge ratio (m/z) normalization, baseline elimination, and conversion of absolute peak height measures to height ratios. After preprocessing, the major difficulty encountered was the extremely large number of variables (1676 m/z values) versus the number of examples (41). Dimensionality reduction was treated as an integral part of the classification process; variable selection was coupled with model construction in a single ten-fold cross-validation loop. We explored different experimental setups involving two peak height representations, two variable selection methods, and six induction algorithms, all on both the original 1676-mass data set and on a prescreened 124-mass data set. Highest predictive accuracies (1-2 off-sample misclassifications) were achieved by a multilayer perceptron and Na?ve Bayes, with the latter displaying more consistent performance (hence greater reliability) over varying experimental conditions. We attempted to identify the most discriminant peaks (proteins) on the basis of scores assigned by the two variable selection methods and by neural network based sensitivity analysis. These three scoring schemes consistently ranked four peaks as the most relevant discriminators: 11683, 1403, 17350 and 66107.     

Spectral quality assessment for high-throughput tandem mass spectrometry proteomics

Current techniques in tandem mass spectrometric analyses of cellular protein contents often produce thousands to tens of thousands of spectra per experiment. This study introduces a new algorithm, named SPEQUAL, which is aimed at automated tandem mass spectral quality assessment. The quality of a given spectrum can be evaluated from three basic components: (i) charge state differentiation, (ii) total signal intensity, and (iii) signal-to-noise estimates. The differentiation between single and multiple precursor charge states (i) provides a binary score for a given spectrum. Components (ii) and (iii) provide partial scores which are subsequently summarized and multiplied by the first score. SPEQUAL was applied to over 10,000 data files derived from almost 3,000 tandem mass spectra, and the results (final cumulative scores) were manually verified. SPEQUAL's performance was determined to have high sensitivity and specificity and low error rates for both spectral quality estimates in general and precursor charge state differentiation in particular. Each of the partial scores is controlled by adjustable thresholds to fine-tune SPEQUAL's performance for different analysis pipelines and instrumentation. This spectral quality assessment tool is intended to act in an advisory role to the researcher, assisting in filtration of thousands of spectra typically produced by high throughput tandem mass spectrometric proteome analyses. Lastly, SPEQUAL was implemented as Java GUI-based and command-line-based interfaces freely available for both academic and industrial researchers.     

Calculating absolute and relative protein abundance from mass spectrometry-based protein expression data

Mass spectrometry (MS)-based shotgun proteomics allows protein identifications even in complex biological samples. Protein abundances can then be estimated from the counts of tandem MS (MS/MS) spectra attributable to each protein, provided one accounts for differential MS detectability of contributing peptides. We developed a method, APEX, which calculates Absolute Protein EXpression levels based upon learned correction factors, MS/MS spectral counts and each protein's probability of correct identification. This protocol describes APEX-based calculations in three parts. (i) Using training data, peptide sequences and their sequence properties, a model is built to estimate MS detectability (O(i)) for any given protein. (ii) Absolute protein abundances are calculated from spectral counts, identification probabilities and the learned O(i)-values. (iii) Simple statistics allow calculation of differential expression in two distinct biological samples, i.e., measuring relative protein abundances. APEX-based protein abundances span 3-4 orders of magnitude and are applicable to mixtures of 100s to 1,000s of proteins.     

Identification of <Emphasis Type="BoldItalic">b</Emphasis>-/<Emphasis Type="BoldItalic">y</Emphasis>-ions in MS/MS spectra using a two stage neural network

Independent of the approach used, the ability to correctly interpret tandem MS data depends on the quality of the original spectra. Even in the case of the highest quality spectra, the majority of spectral peaks can not be reliably interpreted. The accuracy of sequencing algorithms can be improved by filtering out such 'noise' peaks. Preprocessing MS/MS spectra to select informative ion peaks increases accuracy and reduces the processing time. Intuitively, the mix of informative versus non-informative peaks has a direct effect on the quality and size of the resulting candidate peptide search space. As the number of selected peaks increases, the corresponding search space increases exponentially. If we select too few peaks then the ion-ladder interpretation of the spectrum will contain gaps that can only be explained by permutations of combinations of amino acids. This will result in a larger candidate peptide search space and poorer quality candidates. The dependency that peptide sequencing accuracy has on an initial peak selection regime makes this preprocessing step a crucial facet of any approach, whether de novo or not, to MS/MS spectra interpretation.We have developed a novel approach to address this problem. Our approach uses a staged neural network to model ion fragmentation patterns and estimate the posterior probability of each ion type. Our method improves upon other preprocessing techniques and shows a significant reduction in the search space for candidate peptides without sacrificing candidate peptide quality.     

Modeling and characterization of multi-charge mass spectra for peptide sequencing

Peptide sequencing using tandem mass spectrometry data is an important and challenging problem in proteomics. We address the problem of peptide sequencing for multi-charge spectra. Most peptide sequencing algorithms currently consider only charge one or two ions even for higher-charge spectra. We give a characterization of multi-charge spectra by generalizing existing models. Using our models, we analyzed spectra from Global Proteome Machine (GPM) [Craig R, Cortens JP, Beavis RC, J Proteome Res 3:1234-1242, 2004.] (with charges 1-5), Institute for Systems Biology (ISB) [Keller A, Purvine S, Nesvizhskii AI, Stolyar S, Goodlett DR, Kolker E, OMICS 6:207-212, 2002.] and Orbitrap (both with charges 1-3). Our analysis for the GPM dataset shows that higher charge peaks contribute significantly to prediction of the complete peptide. They also help to explain why existing algorithms do not perform well on multi-charge spectra. Based on these analyses, we claim that peptide sequencing algorithms can achieve higher sensitivity results if they also consider higher charge ions. We verify this claim by proposing a de novo sequencing algorithm called the greedy best strong tag (GBST) algorithm that is simple but considers higher charge ions based on our new model. Evaluation on multi-charge spectra shows that our simple GBST algorithm outperforms Lutefisk and PepNovo, especially for the GPM spectra of charge three or more.     

Peptide reranking with protein-peptide correspondence and precursor peak intensity information

Searching tandem mass spectra against a protein database has been a mainstream method for peptide identification. Improving peptide identification results by ranking true Peptide-Spectrum Matches (PSMs) over their false counterparts leads to the development of various reranking algorithms. In peptide reranking, discriminative information is essential to distinguish true PSMs from false PSMs. Generally, most peptide reranking methods obtain discriminative information directly from database search scores or by training machine learning models. Information in the protein database and MS1 spectra (i.e., single stage MS spectra) is ignored. In this paper, we propose to use information in the protein database and MS1 spectra to rerank peptide identification results. To quantitatively analyze their effects to peptide reranking results, three peptide reranking methods are proposed: PPMRanker, PPIRanker, and MIRanker. PPMRanker only uses Protein-Peptide Map (PPM) information from the protein database, PPIRanker only uses Precursor Peak Intensity (PPI) information, and MIRanker employs both PPM information and PPI information. According to our experiments on a standard protein mixture data set, a human data set and a mouse data set, PPMRanker and MIRanker achieve better peptide reranking results than PetideProphet, PeptideProphet+NSP (number of sibling peptides) and a score regularization method SRPI. The source codes of PPMRanker, PPIRanker, and MIRanker, and all supplementary documents are available at our website: http://bioinformatics.ust.hk/pepreranking/. Alternatively, these documents can also be downloaded from: http://sourceforge.net/projects/pepreranking/.