MSClust: a tool for unsupervised mass spectra extraction of chromatography-mass spectrometry ion-wise aligned data

Y. M. Tikunov 0 1 2 3 4 S. Laptenok 0 1 2 3 4 R. D. Hall 0 1 2 3 4 A. Bovy 0 1 2 3 4 R. C. H. de Vos 0 1 2 3 4 0 Y. M. Tikunov Plant Breeding, Wageningen University , 6708 PB Wageningen, The Netherlands 1 Y. M. Tikunov (&) R. D. Hall A. Bovy R. C. H. de Vos Plant Research International , 6700 AA Wageningen, The Netherlands 2 Y. M. Tikunov R. D. Hall A. Bovy R. C. H. de Vos Centre for BioSystems Genomics , 6700 AB Wageningen, The Netherlands 3 R. D. Hall R. C. H. de Vos Netherlands Metabolomics Centre , Einsteinweg 55, 2333 CC Leiden, The Netherlands 4 S. Laptenok Laboratory of Biophysics, Wageningen University , Dreijenlaan 3, 6703 HA Wageningen, The Netherlands Mass peak alignment (ion-wise alignment) has recently become a popular method for unsupervised data analysis in untargeted metabolic profiling. Here we present MSClusta software tool for analysis GC-MS and LC-MS datasets derived from untargeted profiling. MSClust performs data reduction using unsupervised clustering and extraction of putative metabolite mass spectra from ion-wise chromatographic alignment data. The algorithm is based on the subtractive fuzzy clustering method that allows unsupervised determination of a number of metabolites in a data set and can deal with uncertain memberships of mass peaks in overlapping mass spectra. This Availability and implementation MSClust is freely available for non-commercial users at http://www.metalign.nl. - approach is based purely on the actual information present in the data and does not require any prior metabolite knowledge. MSClust can be applied for both GCMS and LCMS alignment data sets. 1 Introduction In both GCMS and LCMS-based metabolomics platforms, untargeted data analysis using unbiased mass peak acquisition followed by their chromatographic alignment, i.e. ion-wise alignment, has become a popular approach for comparative metabolomics. Software tools that can implement this approach, such as MetAlign (Bamba and Fukusaki 2006; Boccard et al. 2010; De Vos et al. 2007; Ducruix et al. 2008; Keurentjes et al. 2006; Lommen 2009; Lommen et al. 2007; Mal et al. 2009; Peters et al. 2009; Rijk et al. 2009; Tikunov et al. 2005; Tikunov et al. 2010; Vorst et al. 2005), MZMine (Katajamaa et al. 2006), or XCMS (Kind et al. 2007; Nordstrom et al. 2006; Smith et al. 2006; Wikoff et al. 2007), are nowadays widely used in metabolomics studies. They are used for primary processing of raw GCMS or LCMS chromatograms (Fig. 1) and they enable a comprehensive comparative analysis of complex metabolic mixtures by aligning quantitative values of individual mass peaks across samples analyzed. Resulting data matrices can be directly subjected to comparative analysis using various statistical tools. However, this approach has a few drawbacks. Firstly, the resulting mass peak alignment matrices are often extremely large with a disproportionate variable-to-sample ratio, as the amount of variables (i.e. detected mass peaks) may reach Fig. 1 A general workflow of a comparative metabolomics data analysis which is based on mass peak alignment approach. MSClust receives a mass peak alignment data matrix of size M 9 S, where M is a number of mass peaks (often tens thousands) aligned across a number of samples profiled S. As the result it produces a reduced data matrix of size C 9 S, where C a number of putative compounds each represented by a single mass peak (normally a few hundred) aligned across the same number of samples S. Besides, it extracts a mass spectra for each of the compounds C, that in case of GCMS data is compatible with the NIST MSSearch compound identification software tens of thousands. Up to 90% of the variables may be redundant, since each metabolite will be represented by a number of different mass peaks, including molecular fragments, adducts, molecular fragments and isotopes thereof. Moreover, this redundancy may vary between profiling platforms and metabolites, depending upon their concentration, ionization efficiency and specific chemical nature. This leads to an unequal representation of metabolites in the dataset and complicates subsequent multivariate or statistical analyses. Secondly, a direct interpretation of the experimental results is hardly possible, because both the structural information of a metabolite, such as a mass spectrum in case of GCMS and in-source fragments in case of LCMS, is not provided directly as a result of the alignment. Previously, we have reported a mass signal correlation analysis approach that can reduce the metabolite signal redundancy in untargeted ion-wise aligned GCMS datasets and to extract mass spectra of individual metabolites without using mass spectral libraries or other structural sources (Tikunov et al. 2005). Here we present a computational implementation of this approachMSClust. In an untargeted metabolomics data analysis workflow it can be placed between the mass peak alignment step and metabolite identification followed by data interpretation. MSClust clusters the aligned mass peaks into reconstructed metabolites, thereby (i) reducing the signal redundancy per metabolite into single representative variables, and (ii) reconstructing original mass spectra, thus providing structural information of the metabolites. This MSClust software tool can be applied to both GCMS and LCMSderived datasets, and for both nominal mass and accurate mass data. The MSClust tool aligns with the Metabolomics Standards Initiative for data processing. 2 Method and implementation The MSClust algorithm aims to remove metabolite signal redundancy in aligned mass peaks tables and to retrieve mass spectral information of metabolites using mass peak clustering. Many clustering methods, e.g. k-means or c-means clustering, self-organizing maps etc., require prior knowledge about a number of clusters in the data. Therefore, these methods cannot be used for chromatography-mass spectrometry data clustering as a number of metabolites is unknown and may vary from tens to hundreds from experiment to experiment. The subtractive fuzzy clustering (Chiu 1994) implemented in the MSClust algorithm allows unsupervised determination of a number of clusters and simultaneous clustering of mass peaks in the mass peak alignment data. The algorithm of MSClust performs clustering of ion-fragments in the dataset that originate from a single metabolite, based on two properties: (i) similarity of chromatography, i.e. retention time span covered by a chromatographic peak of a metabolite, and (ii) quantitative similarity of ion-fragment patterns across a number of samples analyzed. The algorithm performs the following tasks: A number of mass peak clusters (putative metabolites) present in an ion-wise alignment data matrix and cluster centers (centrotype mass peaks) are determined in an unsupervised manner using the potential density (PD) method (Chiu 1994) (Fig. 2A, B) (for detailed explanation of the algorithm see User Manual, Supplemental Data). All mass peaks ar (...truncated)