Hi Desmond,

Majority of the freely available tools, software, and databases from 2014-2015 for your analyses have also been recently summarized/ reviewed by us at Updates in metabolomics tools and resources: 2014–2015 (http://onlinelibrary.wiley.com/doi/10.1002/elps.201500417/abstract), please check out the Table-1 and find your tools!!! : )

Self promotions apart, : )  some of the good UPLC-MS/MS data have been published by Professor Lloyd Sumner (https://scholar.google.com/citations?user=uaN9_BUAAAAJ&hl=en)  and can be a good starting point to see how QC was done for their publications.

I am surprised with the number if metabolites at 1300 (Are they really IDed?) or just m/z' s? If assumed 'identified' then a lot may be "junk" and I or any one interfacing with biological mass spectrometry/ metabolomics would be able to 'weed them out'. Happy to collaborate. Also, counts need to be normalized, scaled, log transformed (or other methods) and so on to see the real effects than counts!

QC wise, (i) you can check for batch variations for example using 'limma' package on R or say normalyzeR or LOESS-N normalization and so on; (ii) internal standard/ QC-pooled runs or even blank runs, based normalization would also tell you the distribution of data/ values, (iii) TIC values and distribution and so on- both pre-processing and statistical tools would let you know the quality of the runs.

I would suggest to run the data through simple WebServer as MetaboAnalyst http://www.metaboanalyst.ca/ or say, WorkflowforMetabolomics: http://workflow4metabolomics.org/ or say XCMSOnline https://xcmsonline.scripps.edu/landing_page.php?pgcontent=mainPage to see the data quality, trends, basic stats, and so on before starting with R and playing with options one by one. Of course bioconductor/R and Python or even Matlab or Galaxy would lead you into a big and different world of analysis of metabolomics data sets.

In our review you can go to the 'holistic tools' section to find many of these press-button software suites to get started and reduce data - in terms of removing "obvious" outliers (say, those which do not fit in PCA at all!) from multivariate analysis tools, which do not have significant changes, IDs which do not make sense (MB Role ID conversion, or say CTS:

Best wishes.

Hope it helps.

Thanks,

Biswa