Dear Luciano Huergo

You should be able to run the 3 functions where IPO generated the pars independently. 'fillPeaks' needs grouping information (because it tries to fill missing peaks based on replicates of the same type).

Please try to provide 'sclass' as an additional parameter to your xcmsSet call. It should be a factor indicating the groups of your samples.

If you process 10 files with 5 replicates of group A and B respectively than 'gl(2,5)' would give you a reasonable input for 'sclass'. You can avoid specifying sclass explicitly when you put your input files into different subfolders (which is what the xcms people prefer I believe). Than xcms should 'guess' the grouping based on the subfolder structure.

Hi Luciano,

your concerns are justified but there might not be an optimal solution. Your main focus should be on avoiding False Positive results, i.e. publishing x metabolites to be differently between your strains/treatments when in fact this would be an artifact of your extraction efficiency.

The Problem:

If you extract/measure twice the amount of sample material of *one* specific sample you will find many signals to be increased/doubled. To make a fair comparison over two or more samples you therefore need to standardize the amount you subject to extraction. Your options for a standardization parameter are cell number (if cells are of similar volume), protein content (if cells are...), genetic markers... However, in a complex experiment (many strains, several treatments) you will not be able to guarantee that there is something consistent enough to be used for normalization.

What you should do:

Try to keep sample amount as constant as possible (OD is fine, counting too). Then monitor potential biases before you do further analyses. As an approximation you can always use the median peak height/size of each sample. Here, your assumption would be that if you extract a large number of analytes, each analyte intensity may differ between samples but on average the amount should be similar. Plot the median peak size of all samples over the replicate groups (group1=strain1, treatment1, replicates1-5, group2=strain1, treatment2, replicates1-5, etc) and run an ANOVA. Most likely you'll find significant differences between groups. Which groups are low, which are high? Is there a good biological explanation for this? Then go on with your analyses. If not, then it is probably a technical artifact leaving you with two options. (1) normalize peak data using median peak size, protein content or something similar, i.e. as described here 10.1007/978-1-4939-7819-9_20 or (2) perform only downstream analyses which are not/less affected by such a potential bias, i.e. investigate metabolite ratios within samples.

Your 13C standards will give you a good hint regarding the technical reproducibility. They will not help you to correct for differences in extraction efficiency (because you add them) but allow you to separate the extraction bias from the remaining downstream processing bias which may be present. If your 13C standards are well correlated with the median I would be surprised (it would mean that your processing and machine performance cause the major technical bias). More likely your standards will be more stable between replicate groups than the median peak size, indicating the need for normalization.

Bests, Jan

Dear Jan

Many thanks for your helpful comments and insights !