Skip to content

pre_process Details

The Pre-processing step performs the LC-MS data processing using XCMS1 (feature finding) and make a correspondence between all MS2 ions and the list of MS1 peaks. It creates one pre-processed MGF file for each sample with all detected MS2 ions enriched with their corresponding MS1 peak dimensions.

Methodology

The pre-process step uses the R package XCMS CentWave algorithm1 to process the LC-MS data and to extract the samples' list of MS1 peaks, the XCMS::PeakGroup algorithm2 to measure the disalignment between the samples, the MSnbase R package3 to read the raw MS/MS data and a modified version of this package function to write to MGF files the pre-processed MS/MS data enriched with the MS1 peak dimensions information. Four fields are added to the header of the pre-processed MGF files to store the MS1 information: RTMIN, RTMAX, PEAK_ID and PEAK_AREA.

The pre-process will enrich the MS2 spectra with a MS1 peak dimension. Each detected MS2 ion's retention time and precursor m/z are matched against its sample list of MS1 peaks. The closest peak that encompasses the spectrum is assigned to it. This match reference is used by the write MGF function to write the peak dimensions information (retention time minimum and maximum, the peak area and the peak ID) in the header of each MS2 ion of the resulting pre-processed MGFs. The peak ID is unique for each MS1 peak, it is given by the CentWave algorithm and has the spectra's sample code of origin added as a suffix. The MS2 retention time is kept as the center of the peak.

Blank samples (in the metadata table SAMPLE_TYPE equals 'blank') receive a different treatment when enriching the spectra. To deal with the baseline problem, e.g., no defined MS1 peak for a given m/z, all MS2 ions of the blank samples are assigned with a baseline peak dimension which encompasses the entire chromatogram from retention time 0 to 1000000 seconds. This allows for more false positive spectra to be removed from the results.

No correspondence between MS1 and MS2

In the matching peak step, depending on the parameter's values used, mostly the peak_width, some peaks can be incorrectly integrated by the CentWave algorithm generating false positive and false negative peaks. This can lead to no match between a MS2 ion and a MS1 peak. The reasons for not finding a peak match can be mostly because of three facts:

  1. bad defined values for the peak_width parameter;
  2. "bad" peaks that do not have a good gaussian shape (e.g., jagged peaks, very close intersecting isomers peaks) can make the CentWave algorithm not work properly and wrongly integrate the peaks (since it is an approximated solution)45;
  3. bad choice of the mz_tolerance or the rt_tolerance parameters values that could be too small to allow a match.

The first and third cases can be modified as needed by the user for each job, and the second case depends on the data quality.

When no MS1 peak match is found for an MS2 spectra, a "fake" peak is created. The mean peak width, provided in the peak_width parameter, is used to simulate the fake peak width centered in the MS2 spectra retention time (RTmean +- mean(peak_width)/2) and the spectra intensity is used as the fake peak area (the clustering will simulate the peak area integration by summing the MS2 spectra intensities). When the peak match occurs because of the rt_tolerance value, e.g., the MS2 ion retention time is outside the integrated MS1 peak width but within a distance less than the rt_tolerance value, the peak width for that spectra is enlarged to encompass the spectra retention time correctly. When more than one match is found, the closest MS1 peak with respect to the peak retention time mean (center of the peak) to the spectra's retention time is chosen as the final match.

When no MS2 spectra receives an integrated MS1 peak, this peak dimensions and quantifications by sample are recorded in a file named "MS1_list_no_MS2.csv" inside the processed data folder. These records will generate a new count file with all MS1 peaks that were not fragmented and thus did not produce a MS2 spectra, and therefore, will be missing in the final clustering counts. This count is aggregated after a clustering job to produce the MS1 count file named with the suffix "_peak_area_MS1.csv", which will add up for the job completeness to try to guarantee that any ion that is detected in the data collection samples will be present in the final counts. These list of not fragmented MS1 peaks will also be used to search for isotopic patterns of multiple charged ions as part of the rules used in the annotation step (Step 7).

Samples Misalignment Suggestion

When max_samples_batch_align > 0, the XCMS::adjustRtime-peakGroups algorithm2 is used to simulate an alignment of some of the samples MS1 peaks and suggest a value for the samples misalignment. The maximum deviation found between the sample's retention time is used to give a suggestion for the retention time tolerance value that should be used in the following workflow commands (in the clustering, clean and merge step, NOT in the pre-process step). It must be noted that the alignment does not change the samples retention time, it only gives a suggestion of the retention time misalignment between the samples.

The samples misalignment suggestion is a very computational consuming step and can possibly not work or crash in some computers with limited memory and/or processing capacity. To try to overcome this difficulty three programming decisions were made. First, the blank samples (metadata column SAMPLE_TYPE equals 'blank') are not used in the misalignment suggestion, mainly because these samples usually have a lot of noise that could lead to a bad behavior of the algorithm. Second, the samples misalignment suggestion of a same data collection batch can be made using only a given number of samples, defined by the parameter max_samples_batch_align, chosen following the not blank samples order in the provided metadata table (the first not blank sample of each data collection batch appearing in the metadata file is always picked). And third, the misalignment suggestion between all samples is always made using only one representative sample of each data collection batch, which is also hard coded to be the first not blank sample of each data collection batch appearing in the metadata file. These behaviors must be taken into account when setting the max_samples_batch_align value and when defining the samples order in the metadata file. The first sample of each data collection batch in the metadata file should be the more related samples between the data collections, e.g. the ones that have the same polarity or bigger chances of having more peaks in common.

The parameters min_fraction, bin_size, bw and max_features should also be tuned to achieve a better misalignment suggestion result. If the misalignment suggestion persists to fail it is up to the user to guess a good value for the retention time tolerances of the following commands, and this value should represent the real misalignment between the samples. The misalignment suggestion can fail if not enough common MS1 peaks are found between the samples, this is a limitation of the used algorithm.

If the misalignment suggestion succeeds, then the user may evaluate to run the following NP³ MS workflow commands with the suggested retention time tolerance values. The user can increase or decrease this tolerance, but we recommend to try to keep it in a range that will prevent joining close by peaks (isomers) or splitting a large peak depending on the common peak_width found in the raw data. It's up to the users to correctly define this tolerance values.

pre_process Diagnostic and Statistics

Finally, a diagnostic is performed to evaluate the pre-process result based on the rate of MS2 spectra without a MS1 peak correspondence (percentage of fake peaks). And then, a suggestion for better peak_width and rt_tolerance pre-process parameters values is performed to help the user improve this step result when needed.

The diagnostic computes the percentage of MS2 spectra without a MS1 m/z range correspondence (named rate_MS2_no_MS1_mz) and the percentage of MS2 spectra without a MS1 retention time range correspondence (named rate_MS2_no_MS1_rt) for each sample and removing blank m/z. If any of these rates is above a cutoff of 5% for a not blank sample, a report is printed informing the most problematic m/zs by sample to facilitate a manual evaluation of these cases by the user. The user should visualize the chromatogram of these m/zs to identify the problem(s), and then, decide to repeat the pre-process step with other parameters values or to accept the result. A descriptive statistics of these rates is also printed to help the user in this decision, when the job have a big number of samples it can be more reasonable to pay more attention to the mean/median values of these rates.

Following this diagnostic, a suggestion for better peak_width and rt_tolerance parameters values is performed based on the peak width distribution of the MS1 list with a MS2 correspondence (not fake peaks) and without blanks (table 'MS1_list_with_MS2.csv' in the pre-process result). The suggestion is computed as follows:

  • Retention time tolerance = median / 2
  • Peak width range (minimum and maximum) = Q25 - 1.5 * IQR, Q75 + 1.5 * IQR
  • The IQR (interquartile range) factors start at 1.5 and are decreased by 0.25 until the peak width suggestions are
  • between the minimum and maximum peak width limit values.

Where the median, Q25, Q75 and IQR values corresponds, respectively, to the median, first quartile (percentile 25th), third quartile (percentile 75th) and interquartile range values of the peak width distribution of the not fake peaks without blanks. Depending on the shape of this distribution, the suggestions can be more or less accurate.

The user should use these suggestions as a guide for choosing better values for the parameters, depending on the diagnostic result and also based on your own analysis/knowlegde of the data (e.g. chromatograms inspection) and the peak width distribution of the real MS1 peaks.

At the end, a diagnostic of possible split peaks is also performed. It works by counting the number of MS1 peaks by m/z present in the MS2 data, without blanks. The number of MS1 peaks indicates the number of possible isomers by m/z. The distribution of the number of m/zs that had a given number of MS1 peaks is stored in the file 'log_number_mzs_by_number_peaks.png' located in the pre-process result. This diagnostic will print a warning if there are more than 5% of the m/zs with more than 30 MS1 peaks (possible isomers), which can indicate that real peaks are being split in smaller peaks. The user should evaluate if this is really an issue for the dataset being used.

References

  1. Tautenhahn, R., Böttcher, C. & Neumann, S. Highly sensitive feature detection for high resolution LC/MS. BMC Bioinformatics 9, 504 (2008). https://doi.org/10.1186/1471-2105-9-504 (XCMS::CentWave Algorithm) ↩↩

  2. Colin A. Smith, Elizabeth J. Want, Grace O'Maille, Ruben Abagyan and Gary Siuzdak. "XCMS: Processing Mass Spectrometry Data for Metabolite Profiling Using Nonlinear Peak Alignment, Matching, and Identification" Anal. Chem. 2006, 78:779-787 (XCMS::PeakGroup and XCMS::adjustRtime-peakGroups algorithms) ↩↩

  3. Gatto L, Lilley KS. MSnbase-an R/Bioconductor package for isobaric tagged mass spectrometry data visualization, processing and quantitation. Bioinformatics. 2012 Jan 15;28(2):288-9. doi: 10.1093/bioinformatics/btr645. Epub 2011 Nov 22. PMID: 22113085 (MSnbase R package) ↩

  4. Owen D. Myers, Susan J. Sumner, Shuzhao Li, Stephen Barnes, and Xiuxia Du. Detailed Investigation and Comparison of the XCMS and MZmine 2 Chromatogram Construction and Chromatographic Peak Detection Methods for Preprocessing Mass Spectrometry Metabolomics Data. Analytical Chemistry 2017 89 (17), 8689-8695 DOI: 10.1021/acs.analchem.7b01069 (XCMS results investigation) ↩

  5. Myers OD, Sumner SJ, Li S, Barnes S, Du X. One Step Forward for Reducing False Positive and False Negative Compound Identifications from Mass Spectrometry Metabolomics Data: New Algorithms for Constructing Extracted Ion Chromatograms and Detecting Chromatographic Peaks. Anal Chem. 2017 Sep 5;89(17):8696-8703. doi: 10.1021/acs.analchem.7b00947. Epub 2017 Aug 17. PMID: 28752754. (XCMS results investigation) ↩