Critical assessment of alignment procedures for LC-MS proteomics and metabolomics measurements.

BACKGROUND

Liquid chromatography coupled to mass spectrometry (LC-MS) has become a prominent tool for the analysis of complex proteomics and metabolomics samples. In many applications multiple LC-MS measurements need to be compared, e. g. to improve reliability or to combine results from different samples in a statistical comparative analysis. As in all physical experiments, LC-MS data are affected by uncertainties, and variability of retention time is encountered in all data sets. It is therefore necessary to estimate and correct the underlying distortions of the retention time axis to search for corresponding compounds in different samples. To this end, a variety of so-called LC-MS map alignment algorithms have been developed during the last four years. Most of these approaches are well documented, but they are usually evaluated on very specific samples only. So far, no publication has been assessing different alignment algorithms using a standard LC-MS sample along with commonly used quality criteria.

RESULTS

We propose two LC-MS proteomics as well as two LC-MS metabolomics data sets that represent typical alignment scenarios. Furthermore, we introduce a new quality measure for the evaluation of LC-MS alignment algorithms. Using the four data sets to compare six freely available alignment algorithms proposed for the alignment of metabolomics and proteomics LC-MS measurements, we found significant differences with respect to alignment quality, running time, and usability in general.

CONCLUSION

The multitude of available alignment methods necessitates the generation of standard data sets and quality measures that allow users as well as developers to benchmark and compare their map alignment tools on a fair basis. Our study represents a first step in this direction. Currently, the installation and evaluation of the "correct" parameter settings can be quite a time-consuming task, and the success of a particular method is still highly dependent on the experience of the user. Therefore, we propose to continue and extend this type of study to a community-wide competition. All data as well as our evaluation scripts are available at http://msbi.ipb-halle.de/msbi/caap.