Feature Ranking and Screening for Class-Imbalanced Metabolomics Data Based on Rank Aggregation Coupled with Re-Balance.

Feature screening is an important and challenging topic in current class-imbalance learning. Most of the existing feature screening algorithms in class-imbalance learning are based on filtering techniques. However, the variable rankings obtained by various filtering techniques are generally different, and this inconsistency among different variable ranking methods is usually ignored in practice. To address this problem, we propose a simple strategy called rank aggregation with re-balance (RAR) for finding key variables from class-imbalanced data. RAR fuses each rank to generate a synthetic rank that takes every ranking into account. The class-imbalanced data are modified via different re-sampling procedures, and RAR is performed in this balanced situation. Five class-imbalanced real datasets and their re-balanced ones are employed to test the RAR's performance, and RAR is compared with several popular feature screening methods. The result shows that RAR is highly competitive and almost better than single filtering screening in terms of several assessing metrics. Performing re-balanced pretreatment is hugely effective in rank aggregation when the data are class-imbalanced.

Establishment of a pattern recognition metabolomics model for the diagnosis of hepatocellular carcinoma.

BACKGROUND: Early diagnosis of hepatocellular carcinoma may help to ensure that patients have a chance for long-term survival; however, currently available biomarkers lack sensitivity and specificity. AIM: To characterize the serum metabolome of hepatocellular carcinoma in order to develop a new metabolomics diagnostic model and identifying novel biomarkers for screening hepatocellular carcinoma based on the pattern recognition method. METHODS: Ultra-performance liquid chromatography-mass spectroscopy was used to characterize the serum metabolome of hepatocellular carcinoma (n = 30) and cirrhosis (n = 29) patients, followed by sequential feature selection combined with linear discriminant analysis to process the multivariate data. RESULTS: The concentrations of most metabolites, including proline, were lower in patients with hepatocellular carcinoma, whereas the hydroxypurine levels were higher in these patients. As ordinary analysis models failed to discriminate hepatocellular carcinoma from cirrhosis, pattern recognition analysis was used to establish a pattern recognition model that included hydroxypurine and proline. The leave-one-out cross-validation accuracy and area under the receiver operating characteristic curve analysis were 95.00% and 0.90 [95% Confidence Interval (CI): 0.81-0.99] for the training set, respectively, and 78.95% and 0.84 (95%CI: 0.67-1.00) for the validation set, respectively. In contrast, for α-fetoprotein, the accuracy and area under the receiver operating characteristic curve were 65.00% and 0.69 (95%CI: 0.52-0.86) for the training set, respectively, and 68.42% and 0.68 (95%CI: 0.41-0.94) for the validation set, respectively. The Z test revealed that the area under the curve of the linear discriminant analysis model was significantly higher than the area under the curve of α-fetoprotein (P < 0.05) in both the training and validation sets. CONCLUSION: Hydroxypurine and proline might be novel biomarkers for hepatocellular carcinoma, and this disease could be diagnosed by the metabolomics model based on pattern recognition.

A Comparison of Sparse Partial Least Squares and Elastic Net in Wavelength Selection on NIR Spectroscopy Data.

Elastic net (Enet) and sparse partial least squares (SPLS) are frequently employed for wavelength selection and model calibration in analysis of near infrared spectroscopy data. Enet and SPLS can perform variable selection and model calibration simultaneously. And they also tend to select wavelength intervals rather than individual wavelengths when the predictors are multicollinear. In this paper, we focus on comparison of Enet and SPLS in interval wavelength selection and model calibration for near infrared spectroscopy data. The results from both simulation and real spectroscopy data show that Enet method tends to select less predictors as key variables than SPLS; thus it gets more parsimony model and brings advantages for model interpretation. SPLS can obtain much lower mean square of prediction error (MSE) than Enet. So SPLS is more suitable when the attention is to get better model fitting accuracy. The above conclusion is still held when coming to performing the strongly correlated NIR spectroscopy data whose predictors present group structures, Enet exhibits more sparse property than SPLS, and the selected predictors (wavelengths) are segmentally successive.

Tuning model parameters in class-imbalanced learning with precision-recall curve.

An issue for class-imbalanced learning is what assessment metric should be employed. So far, precision-recall curve (PRC) as a metric is rarely used in practice as compared with its alternative of receiver operating characteristic (ROC). This study investigates the performance of PRC as the evaluating criterion to address the class-imbalanced data and focuses on the comparison of PRC with ROC. The advantages of PRC over ROC on assessing class-imbalanced data are also investigated and tested on our proposed algorithm by tuning the whole model parameters in simulation studies and real data examples. The result shows that PRC is competitive with ROC as performance measurement for handling class-imbalanced data in tuning the model parameters. PRC can be considered as an alternative but effective assessment for preprocessing (such as variable selection) skewed data and building a classifier in class-imbalanced learning.

Application of Subwindow Factor Analysis and Mass Spectral information for accurate alignment of non-targeted metabolic profiling.

The peak shifts may lead to an incorrect statistical result for nontargeted metabolomics profiling, such as classification and discrimination in pattern recognition. In the paper, a more accurate alignment algorithm is developed based on Subwindow Factor Analysis and Mass Spectral information (SFA-MS). Compared with other methods, this new algorithm aligns the peaks more accurately without changing their shapes, especially for the overlapping peak clusters. To begin, the Continuous Wavelet Transform with Haar wavelet as the mother wavelet (Haar CWT) is used to determine the position and width of peaks. On this basis, the candidate drift points are confirmed by Fast Fourier Transform (FFT) cross correlation. Furthermore, the MS fitting degree of the common components between the reference chromatogram and the raw chromatogram is determined by the Subwindow Factor Analysis (SFA). When the MS information between reference and raw peaks is identical, the corresponding moving points are the optimum shifts. It is remarkable that all the peaks are moved through linear interpolation in the non-peak parts, so that the aligned chromatograms remain unchanged. The SFA-MS algorithm was implemented in the Matlab language and is available as an open source package.

A new strategy to prevent over-fitting in partial least squares models based on model population analysis.

Partial least squares (PLS) is one of the most widely used methods for chemical modeling. However, like many other parameter tunable methods, it has strong tendency of over-fitting. Thus, a crucial step in PLS model building is to select the optimal number of latent variables (nLVs). Cross-validation (CV) is the most popular method for PLS model selection because it selects a model from the perspective of prediction ability. However, a clear minimum of prediction errors may not be obtained in CV which makes the model selection difficult. To solve the problem, we proposed a new strategy for PLS model selection which combines the cross-validated coefficient of determination (Qcv(2)) and model stability (S). S is defined as the stability of PLS regression vectors which is obtained using model population analysis (MPA). The results show that, when a clear maximum of Qcv(2) is not obtained, S can provide additional information of over-fitting and it helps in finding the optimal nLVs. Compared with other regression vector based indictors such as the Euclidean 2-norm (B2), the Durbin Watson statistic (DW) and the jaggedness (J), S is more sensitive to over-fitting. The model selected by our method has both good prediction ability and stability.

Systematic and practical solvent system selection strategy based on the nonrandom two-liquid segment activity coefficient model for real-life counter-current chromatography separation.

Solvent system selection is the first step toward a successful counter-current chromatography (CCC) separation. This paper introduces a systematic and practical solvent system selection strategy based on the nonrandom two-liquid segment activity coefficient (NRTL-SAC) model, which is efficient in predicting the solute partition coefficient. Firstly, the application of the NRTL-SAC method was extended to the ethyl acetate/n-butanol/water and chloroform/methanol/water solvent system families. Moreover, the versatility and predictive capability of the NRTL-SAC method were investigated. The results indicate that the solute molecular parameters identified from hexane/ethyl acetate/methanol/water solvent system family are capable of predicting a large number of partition coefficients in several other different solvent system families. The NRTL-SAC strategy was further validated by successfully separating five components from Salvia plebeian R.Br. We therefore propose that NRTL-SAC is a promising high throughput method for rapid solvent system selection and highly adaptable to screen suitable solvent system for real-life CCC separation.

Identifying bioactive components in natural products through chromatographic fingerprint.

Bioactive component identification is a crucial issue in search for new drug leads. We provide a new strategy to search for bioactive components based on Sure Independence Screening (SIS) and interval PLS (iPLS). The method, which is termed as SIS-iPLS, is not only able to find out the chief bioactive components, but also able to judge how many components should be there responsible for the total bioactivity. The method is totally "data-driven" with no need for prior knowledge about the unknown mixture analyzed, therefore especially suitable for effect-directed work like bioassay-guided fractionation. Two data sets, a synthetic mixture system of twelve components and a suite of Radix Puerariae Lobatae extracts samples, are used to test the identification ability of the SIS-iPLS method.

A novel variable selection approach that iteratively optimizes variable space using weighted binary matrix sampling.

In this study, a new optimization algorithm called the Variable Iterative Space Shrinkage Approach (VISSA) that is based on the idea of model population analysis (MPA) is proposed for variable selection. Unlike most of the existing optimization methods for variable selection, VISSA statistically evaluates the performance of variable space in each step of optimization. Weighted binary matrix sampling (WBMS) is proposed to generate sub-models that span the variable subspace. Two rules are highlighted during the optimization procedure. First, the variable space shrinks in each step. Second, the new variable space outperforms the previous one. The second rule, which is rarely satisfied in most of the existing methods, is the core of the VISSA strategy. Compared with some promising variable selection methods such as competitive adaptive reweighted sampling (CARS), Monte Carlo uninformative variable elimination (MCUVE) and iteratively retaining informative variables (IRIV), VISSA showed better prediction ability for the calibration of NIR data. In addition, VISSA is user-friendly; only a few insensitive parameters are needed, and the program terminates automatically without any additional conditions. The Matlab codes for implementing VISSA are freely available on the website: https://sourceforge.net/projects/multivariateanalysis/files/VISSA/.

Prediction of peptide fragment ion mass spectra by data mining techniques.

Accurate prediction of peptide fragment ion mass spectra is one of the critical factors to guarantee confident peptide identification by protein sequence database search in bottom-up proteomics. In an attempt to accurately and comprehensively predict this type of mass spectra, a framework named MS(2)PBPI is proposed. MS(2)PBPI first extracts fragment ions from large-scale MS/MS spectra data sets according to the peptide fragmentation pathways and uses binary trees to divide the obtained bulky data into tens to more than 1000 regions. For each adequate region, stochastic gradient boosting tree regression model is constructed. By constructing hundreds of these models, MS(2)PBPI is able to predict MS/MS spectra for unmodified and modified peptides with reasonable accuracy. Moreover, high consistency between predicted and experimental MS/MS spectra derived from different ion trap instruments with low and high resolving power is achieved. MS(2)PBPI outperforms existing algorithms MassAnalyzer and PeptideART.

Exploring metabolic syndrome serum profiling based on gas chromatography mass spectrometry and random forest models.

Metabolic syndrome (MetS) is a constellation of the most dangerous heart attack risk factors: diabetes and raised fasting plasma glucose, abdominal obesity, high cholesterol and high blood pressure. Analysis and representation of the variances of metabolic profiles is urgently needed for early diagnosis and treatment of MetS. In current study, we proposed a metabolomics approach for analyzing MetS based on GC-MS profiling and random forest models. The serum samples from healthy controls and MetS patients were characterized by GC-MS. Then, random forest (RF) models were used to visually discriminate the serum changes in MetS based on these GC-MS profiles. Simultaneously, some informative metabolites or potential biomarkers were successfully discovered by means of variable importance ranking in random forest models. The metabolites such as 2-hydroxybutyric acid, inositol and d-glucose, were defined as potential biomarkers to diagnose the MetS. These results obtained by proposed method showed that the combining GC-MS profiling with random forest models was a useful approach to analyze metabolites variances and further screen the potential biomarkers for MetS diagnosis.

Selective of informative metabolites using random forests based on model population analysis.

One of the main goals of metabolomics studies is to discover informative metabolites or biomarkers, which may be used to diagnose diseases and to find out pathology. Sophisticated feature selection approaches are required to extract the information hidden in such complex 'omics' data. In this study, it is proposed a new and robust selective method by combining random forests (RF) with model population analysis (MPA), for selecting informative metabolites from three metabolomic datasets. According to the contribution to the classification accuracy, the metabolites were classified into three kinds: informative, no-informative, and interfering metabolites. Based on the proposed method, some informative metabolites were selected for three datasets; further analyses of these metabolites between healthy and diseased groups were then performed, showing by T-test that the P values for all these selected metabolites were lower than 0.05. Moreover, the informative metabolites identified by the current method were demonstrated to be correlated with the clinical outcome under investigation. The source codes of MPA-RF in Matlab can be freely downloaded from http://code.google.com/p/my-research-list/downloads/list.

Identification of integrin ß1 as a prognostic biomarker for human lung adenocarcinoma using 2D-LC-MS/MS combined with iTRAQ technology.

To discover novel lung adenocarcinoma (AdC) biomarkers, isobaric tags for relative and absolute quantitation (iTRAQ)-tagging combined with 2D-LC-MS/MS analysis was used to identify differentially expressed plasma membrane proteins in lung AdC and paired paraneoplastic normal lung tissues (PNLTs) adjacent to tumors. In this study, significant caveolin-1 downregulation and integrin ß1 upregulation was observed in primary lung AdC vs. PNLT. As there has been no report on the association of integrin ß1 with lung AdC, immunohistochemical staining was performed to detect the expression of integrin ß1 in an independent set of archival tissue specimens including 42 cases of PLNT, 46 cases of without lymph node metastasis primary AdC (non-LNM AdC) and 62 cases of LNM AdC; the correlation of their expression levels with clinicopathological characteristics and clinical outcomes were evaluated. Based on the data, upregulation of integrin ß1 was significantly correlated with advanced clinical stage and lymph node metastasis. Integrin ß1 overexpression was significantly associated with advanced clinical stage (P<0.05), lymph node metastasis (P<0.05), increased relapse rate (P<0.05) and decreased overall survival (P<0.05) in AdCs. Cox regression analysis indicated that integrin ß1 overexpression is an independent prognostic factor. The data suggest that integrin ß1 is a potential biomarker for LNM and prognosis of AdC and integrin ß1 upregulation may play an important role in the pathogenesis of AdC.

Exploring the relationship between 5'AMP-activated protein kinase and markers related to type 2 diabetes mellitus.

The importance of 5'AMP-activated protein kinase (AMPK) in regulating glucose and fatty acid metabolism is increasing. Thus, it is regarded as a new pharmacological target for treatment of obesity, insulin resistance and type 2 diabetes mellitus (T2DM). In order to explore the relationships between AMPK and diabetes mellitus, urines samples from four groups of C57 mice, i.e., the normal male and female C57 mice, female C57-AMPK gene knocked-out mice, and male C57-AMPK gene knocked-out mice, were studied by coupling GC/MS with a powerful machine learning method, random forest. The experimentation has been designed as two steps: firstly, the normal male and female mice were compared with male and female C57-AMPK gene knocked-out mice, respectively; then the differences between male C57-AMPK gene knocked-out mice and female C57-AMPK gene knocked-out mice were further detected. Finally, not only the differences between the normal C57 mice and C57-AMPK gene knocked-out mice were observed, but also the gender-related metabolites differences of the C57-AMPK gene knocked-out mice were obviously visualized. The results obtained with this research demonstrate that combining GC/MS profiling with random forest is a useful approach to analyze metabolites and to screen the potential biomarkers for exploring the relationships between AMPK and diabetes mellitus.

Investigation of scrambled ions in tandem mass spectra, part 2. On the influence of the ions on peptide identification.

A comprehensive investigation was performed to understand the influence of sequence scrambling in peptide ions on peptide identification results. To achieve this, four tandem mass spectrometry datasets with scrambled ions included and with them excluded were analyzed by Crux, X!Tandem, SpectraST, Lutefisk, and PepNovo. While the different algorithms differed in their performance, an increase in the number of correctly identified peptides was generally observed when removing scrambled ions, with the exception of the SpectraST algorithm. However, the variation of the match scores upon removal was unpredictable. Following these investigations, an interpretation was given on how the scrambled ions affect peptide identification. Lastly, a simulated theoretical mass spectral library derived from the NIST peptide Libraries was constructed and searched by SpectraST to study whether scrambled ions in predicted mass spectra could affect peptide identification. Consistent with the peptide library search results, no significant variations for dot product scores as well as peptide identification results were observed when these ions were included in the theoretical MS/MS spectra. From the five adopted algorithms, the SpectraST and Crux provided the most robust results, whereas X!Tandem, PepNovo, and Lutefisk were sensitive to the existence of the scrambled ions, especially the latter two de novo sequencing algorithms.

Investigation of scrambled ions in tandem mass spectra. Part 1. Statistical characterization.

Scrambled ions have become the focus of recent investigations of peptide fragmentation. Here, an investigation of more than 390,000 high quality CID mass spectra is presented to explore the extent of scrambled ions in mass spectra and the possible fragmentation rules during scramble reactions. For the former, scrambled ions generally make up more than 10 % of mass spectra in number, although the abundances are less than 0.1 of the base peak. For the latter, relatively preferential re-opening sites were found for aliphatic residues Ala, Ile, Leu, and other residues such as Met, Gln, Ser, Phe, and Thr, whereas disfavored sites were found for basic residues Arg, Lys, and His, and Trp for both scrambled b and a ions. Similar preferential order in re-opening reaction was found in the reaction of losing internal residues when cleavage occurs at C-terminal side of 20 residues. However, when cleavage occurs at N-terminal side, Glu, Phe, and Trp become the most preferential sites. These results provide a deep insight into cleavage rules during scramble reactions for prediction of peptide mass spectra. Also, an additional investigation of whether scrambled ions could help discriminate false identifications from correct identifications was performed. Probing the number fraction of scrambled ions in falsely and correctly interpreted spectra and analyzing the correlation between scrambled ions and SEQUEST scores XCorr and Sp showed scrambled ions could at some extent help improve the discrimination in singly charged identifications, whereas no improvement was found for multiply charged results.

A novel kernel Fisher discriminant analysis: constructing informative kernel by decision tree ensemble for metabolomics data analysis.

Large amounts of data from high-throughput metabolomics experiments become commonly more and more complex, which brings an enormous amount of challenges to existing statistical modeling. Thus there is a need to develop statistically efficient approach for mining the underlying metabolite information contained by metabolomics data under investigation. In the work, we developed a novel kernel Fisher discriminant analysis (KFDA) algorithm by constructing an informative kernel based on decision tree ensemble. The constructed kernel can effectively encode the similarities of metabolomics samples between informative metabolites/biomarkers in specific parts of the measurement space. Simultaneously, informative metabolites or potential biomarkers can be successfully discovered by variable importance ranking in the process of building kernel. Moreover, KFDA can also deal with nonlinear relationship in the metabolomics data by such a kernel to some extent. Finally, two real metabolomics datasets together with a simulated data were used to demonstrate the performance of the proposed approach through the comparison of different approaches.

Fingerprinting alterations of secondary metabolites of tangerine peels during growth by HPLC-DAD and chemometric methods.

Tangerine peels are herbal materials of two coupled traditional Chinese medicines, Pericarpium Citri Reticulatae (PCR) and Pericarpium Citri Reticulatae Viride (PCRV). In this paper, high-performance liquid chromatographic fingerprints of tangerine peels during growth were firstly measured for deliberately collected 34 samples from three species (Citrus reticulata 'Chachi', Citrus reticulata 'Dahongpao' and Citrus erythrosa Tanaka). After sixteen characteristic components which have similar change trends in the grown process were screened out with the help of heuristic evolving latent projection (HELP) method, score plots of principal component analysis (PCA) successfully presented the grown footprints of tangerine peels. It implied that July might be the best harvest time for PCRV, November and December were better for PCR. Furthermore, hesperidin, nobiletin and tangeretin were screened as chemical markers by loadings of PCA. The HPLC-HELP-PCA strategy has shown its potential in optimization of harvest time and chemical markers' screening, which will have wide perspective in the analysis of "coupled TCMs".

Traditional Chinese medicine and separation science.

Traditional Chinese medicines (TCMs) are getting more and more popular nowadays in the whole world for improving health condition of human beings as well as preventing and healing diseases. TCM is a multi-component system with components mostly unknown, and only a few compounds are responsible for the pharmaceutical and/or toxic effects. The large numbers of other components in the TCM make the screening and analysis of the bioactive components extremely difficult. So, separation and analysis of the desired chemical components in TCM are very important subjects for modernization research of TCM. Thus, many novel separation techniques with significant advantages over conventional methods were introduced and applied to separation and analysis of the chemical constituents in TCM. This review presents just a brief outline of the applications of different separation methods for the isolation and analysis of TCM constituents.

Comparative analysis of volatile components from Clematis species growing in China.

The volatile components between stems and roots and also among five Clematis species from China were studied and analyzed by gas chromatography-mass spectrometry (GC-MS) combined with alternative moving window factor analysis (AMWFA), a new chemometric resolution method. Identification of the compounds was also assisted by comparison of temperature-programmed retention indices (PTRIs) on HP-5MS with authentic samples included in our own laboratory database under construction. A total of 153 different compounds accounting for 86.6-96.5% were identified and significant qualitative and quantitative differences were observed among the samples. The major volatile components in different essential oils from Clematis species were n-hexadecanoic acid and (Z,Z)-9,12-octadecadienoic acid. Our work further demonstrated chemometric resolution techniques upon the two-dimensional data and PTRIs can provide a complementary and convenient method for fast and accurate analysis of complex essential oils.