Time Alignment Algorithms Based on Selected Mass Traces for Complex LC-MS Data

Time alignment of complex LC-MS data remains a challenge in proteomics and metabolomics studies. This work describes modifications of the Dynamic Time Warping (DTW) and the Parametric Time Warping (PTW) algorithms that improve the alignment quality for complex, highly variable LC-MS data sets. Regular DTW or PTW use one-dimensional profiles such as the Total Ion Chromatogram (TIC) or Base Peak Chromatogram (BPC) resulting in correct alignment if the signals have a relatively simple structure. However, when aligning the TICS of chromatograms from complex mixtures with large concentration variability such as serum or urine, both algorithms often lead to misalignment of peaks and thus incorrect comparisons in the subsequent statistical analysis. This is mainly due to the fact that compounds with different m/z values but similar retention times are not considered separately but confounded in the benefit function of the algorithms using only one-dimensional information. Thus, it is necessary to treat the information of different mass traces separately in the warping function to ensure that compounds having the same m/z value and retention time are aligned to each other. The Component Detection Algorithm (CODA) is widely used to calculate the quality of an LC-MS mass trace. By combining CODA with the warping algorithms of DTW or PTW (DTW-CODA or PTW-CODA), we include only high quality mass traces measured by CODA in the benefit function. Our results show that using several CODA selected high quality mass traces in DTW-CODA and PTW-CODA significantly improves the alignment quality of three different, highly complex LC-MS data sets. Moreover, DTW-CODA leads to better preservation of peak shape as compared to the original DTW-TIC algorithm, which often suffers from a substantial peak shape distortion. Our results show that combination of CODA selected mass traces with different time alignment algorithm is a general principle that provide accurate alignment for highly complex samples with large concentration variability.