GPower is an excellent tool for this purpose: http://www.gpower.hhu.de/. But, I might like to discuss more on what is meant by 'power analyses for untargeted metabolomics'. In terms of samples (runs) for reliable calling of features, right ? Thanks.  

You can have a look at this paper 

Data-Driven Sample Size Determination for Metabolic Phenotyping Studies

Benjamin J. Blaise * in Analytical Chemistry

http://pubs.acs.org/doi/abs/10.1021/ac4022314

Dear Biswapriya,

Thanks for your feedback. Yes, for reliable calling of features. In large scale metabolomics studies, it is assumed that since the number of samples are large, there is sufficient power to detect differences in the metabolites. However, some of these studies present differences in over 200-250 different metabolites and some differences are very small but significant. Some of these studies check each metabolite and find if the sample size was adequate to flag the significant difference (at least 80% power). This seems to be a bottom up approach which I am not quite used to. 

Sri, regarding P and N in metabolomics, I have a nice thread to share: http://www.metastars.org/p/94/. But then, yes, in untargetted the numbers goes to 1000s, with majority false positives (adducts, ionization and mass-spec offshoots, over laps, true negatives and so on), it would be impossible to surpass for N ! Are we zeroing on running the same samples again and again until confidence (power) or increasing N! Need help from more stats and -omics guys in this forum ! Biswa