I don't think it is an issue with the conversion of the files as most people suggest on here. We have tested XCMS and compared it to the output generated by Progenesis as well. I played with different settings in order to increase the number of features detected by it. It was always shocking to me that whereas Progenesis could detect tens of thousands of features in a given study, XCMS would come up with less than 10,000. The settings I changed were in the xcmsSet() function : narrower peakwidth() (depends on your LC system, UHPLC/HPLC) and lower ppm values in snthresh() (5 or less). In addition to this setting, it was critical to set the grouping of features in group() to a VERY narrow mzwid window. Even playing around with these settings never got us to as many features as the ones detected by Progenesis and it yielded poor quantitation of low abundance features. The quantitation of robustly detected features ("nice peaks") was pretty comparable between both tools, but when features are low abundance they tend to get clustered together despite having clearly different m/z values (up to 1 amu apart!). In the output of XCMS you can see the m/z windows for given features, some are +/- 1 mu which depending on the mass of the compound could be several thousand ppms. This is an example output for one of the features that I picked randomly:
Correct me if I am wrong, but I interpret this as the measurements for a feature that was measured in a bin of (- 58 ppm to + 21 ppm) around m/z 738.5467?? If you have any feedback or think our analysis of XCMS could be improved somehow, I would love to hear about what people do setting up the mz window in their runs.