Application of Market Basket Analysis for the Visualization of

In this research, our team principally focused on a particular application of market basket analysis (another name for association analysis or frequen...
20 downloads 9 Views 2MB Size
Download PDF
Article pubs.acs.org/ac

Application of Market Basket Analysis for the Visualization of Transaction Data Based on Human Lifestyle and Spectroscopic Measurements Yuka Shiokawa,† Takuma Misawa,†,‡ Yasuhiro Date,†,‡ and Jun Kikuchi*,†,‡,§ †

Graduate School of Medical Life Science, Yokohama City University, 1-7-29 Suehiro-cho, Tsurumi-ku, Yokohama, Kanagawa 230-0045, Japan ‡ RIKEN Center for Sustainable Resource Science, 1-7-22 Suehiro-cho, Tsurumi-ku, Yokohama, Kanagawa 230-0045, Japan § Graduate School of Bioagricultural Sciences, Nagoya University, 1 Furo-cho, Chikusa-ku, Nagoya, Aichi 464-0810, Japan S Supporting Information *

ABSTRACT: With the innovation of high-throughput metabolic profiling methods such as nuclear magnetic resonance (NMR), data mining techniques that can reveal valuable information from substantial data sets are constantly desired in this field. In particular, for the analytical assessment of various human lifestyles, advanced computational methods are ultimately needed. In this study, we applied market basket analysis, which is generally applied in social sciences such as marketing, and used transaction data derived from dietary intake information and urinary chemical data generated using NMR and inductively coupled plasma optical emission spectrometry measurements. The analysis revealed several relationships, such as fish diets with high trimethylamine N-oxide excretion and N-methylnicotinamide excreted at higher levels in the morning and produced from a protein that was consumed one day prior. Therefore, market basket analysis can be applied to metabolic profiling to effectively understand the relationships between metabolites and lifestyle.

I

n metabolomics1 and metabonomics2 studies, nuclear magnetic resonance (NMR) spectroscopy is widely used for comprehensive detection of metabolites because it is an effective and nondestructive analytical method for assessing complex biological mixtures in living organisms.3−8 A typical NMR-based metabolomics/metabonomics approach consists mainly of sample preparation, NMR measurement (usually measured as a one-dimensional 1H NMR spectra, though a recently reported possibility is the fast two-dimensional NMR method),9 signal identification (annotations), and data mining (statistical and multivariate analyses) for capturing relative variations in metabolites and for rapid detection of metabolic biomarkers in biological systems.10−13 Thus, the analytical usefulness and versatility of the NMR-based approach have been supported by the fruits of considerable scientific progress in this regard, e.g., web-based tools and databases for data handling, multivariate analyses, and metabolite identification and annotation, such as MetaboAnalyst,14−16 the open-source package MVAPACK,17 rNMR,18 the human metabolome database (HMDB),19 the biological magnetic resonance data bank (BMRB),20 and 1H (13C) TOCCATA.21,22 With the innovation of high-throughput analytical methods such as NMR, data mining techniques that can reveal valuable information from enormous quantities of data are perpetually desired. Integrated analysis of various data from heterogeneous © 2016 American Chemical Society

measurements, accomplished by data mining methods such as computational science and statistics-based analyses, is one of the keynote approaches in today’s era of “big data management”. Therefore, a number of contemporary studies addressing integrated analyses of NMR-based metabolomics/ metabonomics assessments have been widely reported in the “-omics” fields, e.g., integration of metadata such as genomes and transcriptomes with metabolomes.23,24 In addition, integrated analyses have been performed with other measurement data, such as that obtained from inductively coupled plasma optical emission spectrometry (ICP-OES) and thermogravimetry−differential thermal analysis of various research subjects including human,25 seaweed,26,27 plant,28 soil,29 sediment,30 and aquatic environments.31 Although these research efforts principally focused on integration of numerical data, incorporation of qualitative data such as age, sex, and lifestyle habits was likewise included within the subject analyses; such a balanced approach is deemed key to the perpetual advancement of analytical techniques utilized within this vein of metabolomics and metabonomics. Received: November 4, 2015 Accepted: January 29, 2016 Published: January 29, 2016 2714

DOI: 10.1021/acs.analchem.5b04182 Anal. Chem. 2016, 88, 2714−2719

Article

Analytical Chemistry

NMR Spectroscopy. Collected urinary samples were measured with a Bruker AVANCE II 700 spectrometer (Bruker Biospin, Rheinstetten Germany). For 1H NMR measurements, suppression of water resonance was performed using presaturation during relaxation delay (3 s) and mixing time (10 ms) per a one-dimensional nuclear Overhauser effect (NOE) correlated spectroscopy (NOESY) pulse sequence, in accordance with a previous study.25 The 1H NMR spectra were acquired via 65 536 data points with a spectral width of 14 098 Hz by 32 transients and 4 dummy scans. For annotation and assignment of the detected signals in the 1H NMR spectra, twodimensional (2D) NMR methods such as 1H−1H total correlation spectroscopy (TOCSY), 1H−13C heteronuclear single quantum coherence (HSQC), HSQC−TOCSY, and 2D 1H−1H J-resolved (Jres) NMR spectroscopy were performed on the 700 MHz NMR spectrometer, according to previous study protocols.25,35 In addition, statistical total correlation spectroscopy (STOCSY)36 was also performed using the aforementioned 1H NMR spectra (n = 311) to enhance the annotation of the subject signals. For 1H−1H TOCSY NMR measurement, 512 complex f1 and 2048 complex f2 points were recorded from 16 scans per f1 increment with an interscan delay of 3 s, with 16 dummy scans in spectral widths of 7003 and 14 098 Hz for f1 and f2, respectively. Moreover, the mixing time (D9) was set to 70 ms. For HSQC and HSQC−TOCSY NMR measurements, 256 complex f1 (13C) and 2048 complex f2 (1H) points were recorded from 64 scans per f1 increment in the obtained spectral widths of 31 692 and 14 098 ppm for f1 and f2, respectively. The free induction decays were multiplied by an exponential window function corresponding to 0.3 Hz (f1) and 1.0 Hz (f2) line broadening factors. For 2D Jres NMR measurements, 256 complex f1 and 32 768 complex f2 points were recorded from 8 scans per f1 increment in spectral widths of 14 098 Hz for f2 and 50 Hz for f1. Spectral processing, such as symmetrization and polynomial baseline correction, was performed via the TopSpin software package (Bruker Biospin, Rheinstetten, Germany). Annotation and assignment of metabolites detected in the NMR spectra were performed by the Chenomx NMR Suite Professional software package (version 5.1; Chenomx Inc., Edmonton, Alberta, Canada), the SpinAssign program37 on the PRIMe Web site (http://prime. psc.riken.jp/), the web-based tool SpinCouple38 (http://emar. riken.jp/spincpl), and the Human Metabolome Database39 (http://www.hmdb.ca/). Previous human metabonomic studies were further consulted and referenced as appropriate.40−42 Data Processing. All 1H NMR spectra were converted to numeric data (32 000 data points) by TopSpin software. The numeric data were normalized by probabilistic quotient normalization43 with the package “mQTL” (https://cran.rproject.org/web/packages/mQTL/index.html) by Revolution R Open (RRO) (8.0.1 beta 64 bit) and then aligned with an inhouse edited version of icoshift44 by Matlab R2015b (MathWorks Japan, Tokyo, Japan) due to nonuniformity caused by variable pH and/or concentrations of human samples. For the aligned spectra, signal peak designations were performed on RRO software to eliminate noise signals in the NMR spectra. The peak data were used for the market basket analysis described below. ICP-OES. The collected urinary samples were measured by ICP-OES (SPS5510, SII Nano-Technology, Chiba, Japan) for evaluation of output minerals, according to previous studies.25,45

In this research, our team principally focused on a particular application of market basket analysis (another name for association analysis or frequent itemset mining), which is capable of integration using quantitative and qualitative data and is often used in marketing analysis, to metabolomics and metabonomics studies. Market basket analysis is based on the concept of finding the co-occurring event in a data set. This analysis is applied in multiple fields, including biology,32,33 with the development of several algorithms. One of the more famous algorithms is “apriori”, which was developed in 1993.34 The current study describes an analytical strategy incorporated within the market basket analysis for capturing the relationship between human lifestyle habits and metabolic responses (Figure 1). In this strategy, human urinary profiles

Figure 1. Analytical flow of this study. The study used variables composed of inputs for humans, such as individual, dietary intake, and time of urinary excretion, and outputs from humans, such as urinary metabolites and minerals, for computation of an integrated analysis. The obtained NMR spectra were preprocessed by peak alignments and peak pickings. All input and output data for individual humans were processed by ranking for conversion to applicable forms for the subject market basket analysis. The data were moreover converted to digitized form, followed by integration via the market basket analysis.

measured by NMR and ICP-OES, as well as nutritional/lifestyle information derived from daily dietary intakes, were preprocessed by ranking and digitalization. This was then followed by integration and evaluation via market basket analysis to reveal input−output responses in humans (Figure 1).

■

MATERIALS AND METHODS Human Subjects and Sample Collections. All human experiments were approved by the human ethical committees of RIKEN Yokohama Research Institute and Yokohama City University. All human volunteers provided informed consent prior to the commencement of research activities. Urinary samples (n = 311) were provided by eight human volunteers and collected in accordance with protocols established in a previous study.25 Nutritional (dietary) information was derived from previous data with additional calculation of information about fish intake of the diet using Excel Eiyo-kun (Kenpakusha, Tokyo, Japan) according to a previous study.25 2715

DOI: 10.1021/acs.analchem.5b04182 Anal. Chem. 2016, 88, 2714−2719

Article

Analytical Chemistry Market Basket Analysis. The numeric data from NMR, ICP-OES, and nutrients were converted to “high” and “low” ranked data. The “high” or “low” ranked data were defined as the top 25% or the bottom 25% of all measured values in each metabolite/mineral/nutrient. The obtained data, tagged by the ranked information (“high”, “low”, or “null”), were integrated with respect to each sample, with each sample ultimately being associated with the time of urine specimen collection. The subject time information was either categorized as “time 1” (morning, from 5:00 a.m. to 12:00 p.m.), “time 2” (afternoon, from 12:00 p.m. to 6:00 p.m.), or “time 3” (evening, from 6:00 p.m. to 12:00 a.m.). Moreover, each sample was also associated with nutrient information that was consumed by the human subject on the day prior. The obtained 574 variables from metabolites, minerals, nutrients, and personal information were ultimately integrated by market basket analysis. Association rules were determined using criterion values of support, confidence, and lift. An association rule is thus expressed in the form X ⇒ Y, where X ∩ Y = ϕ. Let X be a set of some variables in I, I = {i1, i2, ···, im} be a set of all possible variables, likewise Y be a set of some variables in I. Support is the probability of X and Y co-occurring in the transaction data set:

market basket analysis. For metabolite annotations of the signals detected in the 1H NMR spectra, STOCSY analysis and 2D NMR measurements such as TOCSY, HSQC, HSQC− TOCSY, and 2D Jres NMR were ultimately performed (Figure 2 and Figure S1). From these measurements and analyses, the

support(X ⇒ Y ) = P(X ∩ Y )

The conf idence of the rule X ⇒ Y is the conditional probability of observing Y given that X is present in a transaction: confidence(X ⇒ Y ) =

P(X ∩ Y ) P(X )

The lift of the rule X ⇒ Y is the ratio of the support if X and Y are independent. lift(X ⇒ Y ) =

P(X ∩ Y ) P(X )P(Y )

Figure 2. Expansion of urinary 1H NMR spectrum (top), 2D Jres NMR spectrum (middle), and HSQC NMR spectrum (bottom) at 2.90−4.10 ppm. Numbers highlighted along with 1H NMR signals are annotated metabolites listed in Table S1.

Therefore, higher lift values imply a high probability of event Y in the case of condition X. Since lift values 1”. The association network was depicted via the use of the Cytoscape program47 (http://www.cytoscape.org/).

detected signals were annotated (assigned) and summarized in Table S1. A total of 574 variables were used for the market basket analysis; these variables comprised three sample data points (time of urinary excretion), the 424 peaks detected in the NMR spectra, 12 minerals detected in ICP-OES measurements (Figure S2), and 135 dietary nutrients that were consumed the day prior. Association Rule Determination. All association rules were comprehensively visualized in the functions of support, confidence, and the number of rules (Figure S3). By applying the threshold values of support and confidence (yellow line in Figure S3), 510 531 association rules were observed. Among these rules, 4441 rules, input (nutrients and diet) or output (metabolites and minerals) → output (metabolites and minerals) → time 1 (morning), were first extracted as the “important” rules (Figure 3A). From the selected association rules, the important association rules were eventually extracted to satisfy requirements for the rules themselves, i.e., input (nutrients and diets) → output (metabolites and minerals) → time 1 (morning). Moreover, to eliminate the low levels of importance from the association rules, the rules were filtered by a threshold value of 1.2 for lift, resulting in 1307 association rules that were revealed to have high levels of importance

■

RESULTS AND DISCUSSION Data Preprocessing and Metabolite Annotation. This study focused on the advancement of analytical strategies that incorporate association analyses into human input−output response evaluations (Figure 1). To this end, nutritional and lifestyle information derived from daily dietary intakes was provided by human volunteers, with metabolic and mineral profiles in collected human urine samples evaluated by 1H NMR spectra and ICP-OES measurements, respectively. For the 1H NMR spectra, peak alignments and peak determinations were performed via the preprocessing of NMR data for the 2716

DOI: 10.1021/acs.analchem.5b04182 Anal. Chem. 2016, 88, 2714−2719

Article

Analytical Chemistry

Figure 3. Relationship between the number of rules and functions of support and confidence (A) and function of lift (B) using the variables of interest, i.e., only the relationships among input nutrients/diet and output metabolites/minerals which were screened by the analysis shown in Figure S3. Each number of rules was calculated by changing support and confidence values. The yellow line indicates thresholds of support and confidence. The red arrow indicates a lift value of 1.2, the filtering threshold for rules.

Figure 4. Part of the association network at time 1. High levels of TMAO (A) and N-methylnicotinamide (B) with dietary nutrition that was consumed a day prior are shown as association networks. The red, purple, yellow, green, blue, orange, and gray colors indicate output metabolites and minerals in urine, input proteins, fats, carbohydrates, minerals, vitamins, and other nutritional information derived from diets, respectively. “H_” indicates a high level of material, and “L_” indicates a low level of material.

excreted in the morning was related to the intake of fish with high levels of polyunsaturated and n-3 polyunsaturated fatty acids consumed the day prior (Figure 4A and Figure S5). It has been reported that high levels of TMAO excreted in urine are caused by high meat consumption in the diet compared to a low-meat or vegetarian diet.48 In addition to this, fish is a good source of TMAO, as described in a previous report.49 We had previously detected a high-intensity signal of TMAO in the water-soluble components of fish by 1H NMR measurements.50 Therefore, high TMAO excretions in urine might be directly derived from fish components in the diet. Thus, TMAO excretions in urine are heavily influenced by ingestion of fish. Moreover, the association network revealed that N-methyl-

(Figure 3B). It was these important association rules that were ultimately used for the following association network analysis. Association Network Analysis. Association network analysis was performed to find relationships between nutritional information on what was consumed a day prior and metabolites/minerals in resulting urine. The association network enabled the research team to detect comprehensive associations and to visualize input−output responses, i.e., to depict as a network diagram subject dietary intakes (nutritional information) followed by associated metabolic/mineral urinary excretion on the morning of the following day (Figure S4B). From the association network analysis, for example, it was found that a high level of trimethylamine N-oxide (TMAO) 2717

DOI: 10.1021/acs.analchem.5b04182 Anal. Chem. 2016, 88, 2714−2719

Analytical Chemistry

■

nicotinamide (NMNA), detected at high levels within urinary excretions, was associated with several dietary nutrients which were consumed the day prior (Figure 4B). NMNA is a metabolite produced from nicotinamide adenine dinucleotide (NAD) in the ATP production process. This finding suggested that increased urinary NMNA concentrations were caused by the metabolic conversion of nicotinamide to NMNA, invoked by the inhibition of nicotinamide phosphoribosyltransferase activity when NAD pools were affected by decreased energy consumption while sleeping.51 In addition, the association network revealed that high levels of urinary NMNA were correlated to high levels of amino acids and proteins in calculated nutritional data. The result was considered to be consistent with the well-known fact that NAD is produced from niacin and tryptophan, suggesting that the validity of the association network analysis is altogether substantiated. Taken together, this study’s advanced analytical strategy using market basket analysis successfully enabled the research team to capture and interpret metabolic responses caused by human lifestyle habits (such as mealtime regimens and unbalanced diets) into the fields of metabolomics and metabonomics. Such an analytical strategy is thus a versatile and useful tool for evaluation and characterization of the metabolic variations and dynamics in not only biological systems but also environmental ecosystems, i.e., the subject method should be applicable to environmental studies such as environmental metabolomics.52 Moreover, the strategy should be useful not only for NMR-based approaches but also for mass spectrometry (MS)-based metabolomics studies, as well as other univariate and multivariate data obtained from various physicochemical measurement instruments. In future assessments, the research team will apply this analytical strategy to projects on environmental conservation in the quest to integrate multiple data points from various instruments and to further obtain biological knowledge in terms of metabolic variations caused by organism−environment interactions.

AUTHOR INFORMATION

Corresponding Author

*Phone: +81455039439. Fax: +81455039489. E-mail: jun. [email protected]. Author Contributions

The manuscript was written with equal contributions from all of the authors. All the authors have approved the final version of the manuscript. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This research was supported in part by a Grants-in-Aid for Scientific Research (Grant No. 25513012) to J.K. and by the Council for Science, Technology and Innovation (CSTI), Cross-ministerial Strategic Innovation Promotion Program (SIP) “Technologies for creating next-generation agriculture, forestry and fisheries”, funded by the Bio-oriented Technology Research Advancement Institution (NARO).

■

REFERENCES

(1) Goodacre, R.; Vaidyanathan, S.; Dunn, W. B.; Harrigan, G. G.; Kell, D. B. Trends Biotechnol. 2004, 22, 245−252. (2) Nicholson, J. K.; Connelly, J.; Lindon, J. C.; Holmes, E. Nat. Rev. Drug Discovery 2002, 1, 153−161. (3) Torregrossa, L.; Shintu, L.; Nambiath Chandran, J.; Tintaru, A.; Ugolini, C.; Magalhães, A.; Basolo, F.; Miccoli, P.; Caldarelli, S. J. Proteome Res. 2012, 11, 3317−3325. (4) An, Y. J.; Xu, W. J.; Jin, X.; Wen, H.; Kim, H.; Lee, J.; Park, S. ACS Chem. Biol. 2012, 7, 2012−2018. (5) Halouska, S.; Fenton, R. J.; Barletta, R. G.; Powers, R. ACS Chem. Biol. 2012, 7, 166−171. (6) Bingol, K.; Zhang, F.; Bruschweiler-Li, L.; Bruschweiler, R. Anal. Chem. 2013, 85, 6414−6420. (7) Andre, M.; Dumez, J. N.; Rezig, L.; Shintu, L.; Piotto, M.; Caldarelli, S. Anal. Chem. 2014, 86, 10749−10754. (8) Kang, J.; Choi, M.; Kang, S.; Kwon, H.; Wen, H.; Lee, C. H.; Park, M.; Wiklund, S.; Kim, H. J.; Kwon, S. W.; Park, S. J. Agric. Food Chem. 2008, 56, 11589−11595. (9) Guennec, A. L.; Giraudeau, P.; Caldarelli, S. Anal. Chem. 2014, 86, 5946−5954. (10) Beckonert, O.; Keun, H. C.; Ebbels, T. M.; Bundy, J.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Nat. Protoc. 2007, 2, 2692−2703. (11) Powers, R. J. Med. Chem. 2014, 57, 5860−5870. (12) Wen, H.; An, Y. J.; Xu, W. J.; Kang, K. W.; Park, S. Angew. Chem., Int. Ed. 2015, 54, 5374−5377. (13) Tranchida, F.; Shintu, L.; Rakotoniaina, Z.; Tchiakpe, L.; Deyris, V.; Hiol, A.; Caldarelli, S. PLoS One 2015, 10, e0135948. (14) Xia, J. G.; Mandal, R.; Sinelnikov, I. V.; Broadhurst, D.; Wishart, D. S. Nucleic Acids Res. 2012, 40, W127−W133. (15) Xia, J. G.; Psychogios, N.; Young, N.; Wishart, D. S. Nucleic Acids Res. 2009, 37, W652−W660. (16) Xia, J. G.; Sinelnikov, I. V.; Han, B.; Wishart, D. S. Nucleic Acids Res. 2015, 43, W251−W257. (17) Worley, B.; Powers, R. ACS Chem. Biol. 2014, 9, 1138−1144. (18) Lewis, I. A.; Schommer, S. C.; Markley, J. L. Magn. Reson. Chem. 2009, 47, S123−S126. (19) Wishart, D. S.; Tzur, D.; Knox, C.; Eisner, R.; Guo, A. C.; Young, N.; Cheng, D.; Jewell, K.; Arndt, D.; Sawhney, S.; Fung, C.; Nikolai, L.; Lewis, M.; Coutouly, M. A.; Forsythe, I.; Tang, P.; Shrivastava, S.; Jeroncic, K.; Stothard, P.; Amegbey, G.; Block, D.; Hau, D. D.; Wagner, J.; Miniaci, J.; Clements, M.; Gebremedhin, M.; Guo, N.; Zhang, Y.; Duggan, G. E.; Macinnis, G. D.; Weljie, A. M.; Dowlatabadi, R.; Bamforth, F.; Clive, D.; Greiner, R.; Li, L.; Marrie, T.; Sykes, B. D.; Vogel, H. J.; Querengesser, L. Nucleic Acids Res. 2007, 35, D521−526.

■

CONCLUSIONS Market basket analysis is a robust method that can accurately define relationships between various modes of quantitative and qualitative data. The method has been conventionally utilized in social sciences such as marketing and has not previously been implemented for use in metabolomics or metabonomics. In this study, we have developed an analytical strategy using market basket analysis to capture metabolic responses caused by human lifestyle habits. Per the data processed via the subject market basket analysis, we succeeded in revealing several relationships between urinary metabolites and nutrition based on food that was consumed a day prior to the time of urinary excretion. This analytical strategy is hence deemed a versatile and useful approach for evaluating and characterizing metabolic variations/dynamics in biological and environmental ecosystems.

■

Article

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.analchem.5b04182. Five figures, including NMR spectra, association network, and a list of annotated metabolites (PDF) 2718

DOI: 10.1021/acs.analchem.5b04182 Anal. Chem. 2016, 88, 2714−2719

Article

Analytical Chemistry (20) Ulrich, E. L.; Akutsu, H.; Doreleijers, J. F.; Harano, Y.; Ioannidis, Y. E.; Lin, J.; Livny, M.; Mading, S.; Maziuk, D.; Miller, Z.; Nakatani, E.; Schulte, C. F.; Tolmie, D. E.; Kent Wenger, R.; Yao, H.; Markley, J. L. Nucleic Acids Res. 2008, 36, D402−D408. (21) Bingol, K.; Bruschweiler-Li, L.; Li, D. W.; Bruschweiler, R. Anal. Chem. 2014, 86, 5494−5501. (22) Bingol, K.; Zhang, F.; Bruschweiler-Li, L.; Bruschweiler, R. Anal. Chem. 2012, 84, 9395−9401. (23) Mochida, K.; Furuta, T.; Ebana, K.; Shinozaki, K.; Kikuchi, J. BMC Genomics 2009, 10, 568. (24) Ogata, Y.; Chikayama, E.; Morioka, Y.; Everroad, R. C.; Shino, A.; Matsushima, A.; Haruna, H.; Moriya, S.; Toyoda, T.; Kikuchi, J. PLoS One 2012, 7, e30263. (25) Misawa, T.; Date, Y.; Kikuchi, J. J. Proteome Res. 2015, 14, 1526−1534. (26) Ito, K.; Sakata, K.; Date, Y.; Kikuchi, J. Anal. Chem. 2014, 86, 1098−1105. (27) Wei, F.; Ito, K.; Sakata, K.; Date, Y.; Kikuchi, J. Anal. Chem. 2015, 87, 2819−2826. (28) Ogura, T.; Date, Y.; Kikuchi, J. PLoS One 2013, 8, e66919. (29) Ogura, T.; Date, Y.; Tsuboi, Y.; Kikuchi, J. ACS Chem. Biol. 2015, 10, 1908−1915. (30) Asakura, T.; Date, Y.; Kikuchi, J. Anal. Chem. 2014, 86, 5425− 5432. (31) Ogawa, D. M.; Moriya, S.; Tsuboi, Y.; Date, Y.; Prieto-da-Silva, A. R.; Radis-Baptista, G.; Yamane, T.; Kikuchi, J. PLoS One 2014, 9, e110723. (32) Chen, S. C.; Tsai, T. H.; Chung, C. H.; Li, W. H. BMC Genomics 2015, 16, 786. (33) Naulaerts, S.; Meysman, P.; Bittremieux, W.; Vu, T. N.; Vanden Berghe, W.; Goethals, B.; Laukens, K. Briefings Bioinf. 2015, 16, 216− 231. (34) Agrawal, R.; Imielinski, T.; Swami, A. ACM SIGMOD 1993, 22, 207−216. (35) Date, Y.; Iikura, T.; Yamazawa, A.; Moriya, S.; Kikuchi, J. J. Proteome Res. 2012, 11, 5602−5610. (36) Cloarec, O.; Dumas, M. E.; Craig, A.; Barton, R. H.; Trygg, J.; Hudson, J.; Blancher, C.; Gauguier, D.; Lindon, J. C.; Holmes, E.; Nicholson, J. Anal. Chem. 2005, 77, 1282−1289. (37) Chikayama, E.; Sekiyama, Y.; Okamoto, M.; Nakanishi, Y.; Tsuboi, Y.; Akiyama, K.; Saito, K.; Shinozaki, K.; Kikuchi, J. Anal. Chem. 2010, 82, 1653−1658. (38) Kikuchi, J.; Tsuboi, Y.; Komatsu, K.; Gomi, M.; Chikayama, E.; Date, Y. Anal. Chem. 2016, 88, 659−665. (39) Wishart, D. S.; Jewison, T.; Guo, A. C.; Wilson, M.; Knox, C.; Liu, Y.; Djoumbou, Y.; Mandal, R.; Aziat, F.; Dong, E.; Bouatra, S.; Sinelnikov, I.; Arndt, D.; Xia, J.; Liu, P.; Yallou, F.; Bjorndahl, T.; Perez-Pineiro, R.; Eisner, R.; Allen, F.; Neveu, V.; Greiner, R.; Scalbert, A. Nucleic Acids Res. 2013, 41, D801−807. (40) Bouatra, S.; Aziat, F.; Mandal, R.; Guo, A. C.; Wilson, M. R.; Knox, C.; Bjorndahl, T. C.; Krishnamurthy, R.; Saleem, F.; Liu, P.; Dame, Z. T.; Poelzer, J.; Huynh, J.; Yallou, F. S.; Psychogios, N.; Dong, E.; Bogumil, R.; Roehring, C.; Wishart, D. S. PLoS One 2013, 8, e73076. (41) Diaz, S. O.; Barros, A. S.; Goodfellow, B. J.; Duarte, I. F.; Carreira, I. M.; Galhano, E.; Pita, C.; Almeida, M. d. C.; Gil, A. M. J. Proteome Res. 2013, 12, 969−979. (42) Vazquez-Fresno, R.; Llorach, R.; Alcaro, F.; Rodriguez, M. A.; Vinaixa, M.; Chiva-Blanch, G.; Estruch, R.; Correig, X.; AndresLacueva, C. Electrophoresis 2012, 33, 2345−2354. (43) Dieterle, F.; Ross, A.; Schlotterbeck, G.; Senn, H. Anal. Chem. 2006, 78, 4281−4290. (44) Savorani, F.; Tomasi, G.; Engelsen, S. B. J. Magn. Reson. 2010, 202, 190−202. (45) Date, Y.; Nakanishi, Y.; Fukuda, S.; Nuijima, Y.; Kato, T.; Umehara, M.; Ohno, H.; Kikuchi, J. Food Chem. 2014, 152, 251−260. (46) Michael, H.; Sundheer, C.; Kurt, H.; Christian, B. J. Mach. Learn. Res. 2011, 12, 2021−2025.

(47) Shannon, P.; Markiel, A.; Ozier, O.; Baliga, N. S.; Wang, J. T.; Ramage, D.; Amin, N.; Schwikowski, B.; Ideker, T. Genome Res. 2003, 13, 2498−2504. (48) Stella, C.; Beckwith-Hall, B.; Cloarec, O.; Holmes, E.; Lindon, J. C.; Powell, J.; van der Ouderaa, F.; Bingham, S.; Cross, A. J.; Nicholson, J. K. J. Proteome Res. 2006, 5, 2780−2788. (49) Heinzmann, S. S.; Merrifield, C. A.; Rezzi, S.; Kochhar, S.; Lindon, J. C.; Holmes, E.; Nicholson, J. K. J. Proteome Res. 2012, 11, 643−655. (50) Yoshida, S.; Date, Y.; Akama, M.; Kikuchi, J. Sci. Rep. 2014, 4, 7005. (51) Okamoto, H.; Ishikawa, A.; Yoshitake, Y.; Kodama, N.; Nishimuta, M.; Fukuwatari, T.; Shibata, K. Am. J. Clin. Nutr. 2003, 77, 406−410. (52) Bundy, J. G.; Davey, M. P.; Viant, M. R. Metabolomics 2009, 5, 3−21.

2719

DOI: 10.1021/acs.analchem.5b04182 Anal. Chem. 2016, 88, 2714−2719