# RMassBank: The workflow by example In RMassBank: Workflow to process tandem MS files and build MassBank records

options(width=74)


# Introduction

RMassBank is a two-part computational mass spectrometry workflow:

• In the first step, MSMS spectra of compounds are extracted from raw LC-MS data files, the MSMS spectra are recalibrated using assigned fragment formulas, and effectively denoised by using only annotated peaks (plus peaks which can be manually added.)
• In the second step, the processed, recalibrated, cleaned data is prepared for submission to a MassBank database. Compounds are first automatically annotated using information from the Chemical Translation Service (CTS). After manually checking and fixing the annotations, the information is compiled together with the spectral data into MassBank records, which can then be uploaded to a MassBank database.

This vignette describes basic usage with the standard workflow. The package is flexible and allows for different advanced use cases. Examples of specialized applications of RMassBank are available at the RMassBank message board hosted by the Metabolomics-Forum: (http://www.metabolomics-forum.com/viewforum.php?f=29).

The library is available from Bioconductor ((http://www.bioconductor.org)). In addition to the library itself, it is recommended to install the OpenBabel chemical toolkit, available from (http://www.openbabel.org) for various platforms (or via Linux package distribution systems).

The library is loaded as follows

library(RMassBank)


The data used in the following example is available as a package RMassBankData, which must be installed separately and is loaded using

library(RMassBankData)


# Input files

## LC/MS data

RMassBank handles high-resolution LC/MS spectra in mzML format in centroid^[The term "centroid" here refers to any kind of data which are not in profile mode, i.e. don't have continuous m/z data. It does not refer to the (mathematical) centroid peak, i.e. the area-weighted mass peak.] or in profile mode.

Data in the examples was acquired using an LTQ Orbitrap XL instrument in profile mode, and converted from profile-mode RAW into centroid-mode mzML using MSConvertGUI from ProteoWizard. The settings were as shown in the screenshot below (note the "Peak Picking" filter.)

Fig. 1: ProteoWiz settings for conversion to mzML

In the standard workflow, the file names are used to identify a compound: file names must be in the format xxxxxxxx_1234_xxx.mzXML, where the xxx parts denote anything and the 1234 part denotes the compound ID in the compound list (see below). Advanced and alternative uses can be implemented; consult the implementation of msms_workflow and findMsMsHR for more information.

## Compound list

A compound list in CSV format is required to identify all compounds unambiguously. The CSV file is required to have at least the following columns, which are used for further processing and must be named correctly (but present in any order): ID, Name, SMILES, RT, CAS. The columns ID and SMILES must be filled, the other columns must be present in the file but do not need to be filled. ID specifies an (arbitrary) numeric ID code which must be < 4 digits long; SMILES specifies a SMILES code with the chemical structure of the compound (and is used to extract the molecular formula, for calculation of molecular masses, for database searching in CTS etc.) Although the columns Name, RT, CAS have to be present, the information in the columns is only used if the cells are filled. RT, if present, specifies the retention time (in minutes; $\pm$ a window specified in the RMassBank options, see below) where a LC/MS file is searched for the compound spectra. CAS and Name are used as additional information while retrieving annotations from CTS. The compound list doesn't have to be ordered in any particular way. It can contain large numbers of compounds, even compounds which will not be actively used by the script (Note: Unused compounds don't require a SMILES code, since they will not be accessed.)

An example list is provided with the RMassBankData package, and can be copied into a local folder, viewed and edited:

file.copy(system.file("list/NarcoticsDataset.csv",
package="RMassBankData"), "./Compoundlist.csv")


## Settings

A number of different settings influence RMassBank. They are partly parameters for data processing and partly constants used for annotation.

A settings template file, to be edited by hand, can be generated using

RmbSettingsTemplate("mysettings.ini")


where mysettings.ini is the file that will be generated. This file should then be edited. Important settings are:

• deprofile: Whether to use a deprofiling algorithm to work with profile-mode data. Default is NA for use with centroid-mode data. Allowed settings for profile-mode data include deprofile.fwhm (full-width half-maximum algorithm), deprofile.spline (cubic spline algorithm), deprofile.localmax (local maximum). See the respective help pages for detailed information.
• rtMargin: The deviation allowed for retention times (in minutes) when extracting spectra from raw data files.
• rtShift: The systematic retention time shift (in minutes) in the LC-MS data compared to the values in the compound list.
• babeldir: The directory pointing to the OpenBabel binaries.
• use_version: which MassBank data format to use. The default is the newer version 2; alternatively, the (deprecated) version 1 can be specified for MassBank servers running old versions of the server software.
• use_rean_peaks: Whether or not peaks from reanalysis should be used (see below for details.)
• add_annotation: Whether or not fragments should be annotated with the (tentative) molecular formula in MassBank records.
• annotations: A list of annotation data used in the MassBank records.
• authors, copyright, publication, license, instrument, instrument_type, compound_class: values for the corresponding MassBank fields
• confidence_comment: A commentary field about "compound confidence" which is added like "COMMENT: CONFIDENCE standard compound" in the MassBank record.
• internal_id_fieldname: The name for an internal ID field in the MassBank record where to store the compound ID (in the compound list). For internal_id_fieldname = "MY_ID", the ID will be stored like "COMMENT: MY_ID 1234".
• entry_prefix: The (2-letter) prefix for MassBank accession IDs.
• ms_type, ionization, lc_*: Annotations for the LC and MS information fields in the MassBank records.
• ms_dataprocessing: Tags added to describe the data processing. In addition to the tags specified here, MS\$DATA_PROCESSING: WHOLE RMassBank will be added (corresponding to a list("WHOLE" = "RMassBank") entry for this option.) • annotator: For advanced users: option to select your own custom annotator. Check ?annotator.default and the source code for details. • spectraList: The list of data-dependent scans triggered by a MS1 scan in their order; used for annotation of MassBank records. See the template file for description. • accessionBuilderType: A string (either "standard", "simple" or "selfDefined") to determine how to generate MassBank record accession numbers (optional, default: "standard"). RMassBank generates an accession number for each record. The structure and generation of this number varies based on accessionBuilderType. • "standard": 2-letter + 6-digit accession numbers are generated. The 2-letter code is defined by annotations$entry_prefix, the first four digits are given by the compound ID. The last two digits are generated from the position of the spectrum in spectraList and the shift defined in accessionNumberShifts for the selected ion type (Example: the compound with ID 2112, processed in "pNa" mode ([M+Na]+), will have accession numbers XX211233, XX211234 ... etc in for the first, second... spectrum in the data-dependent scan, if the "pNa" shift is set to 32.)
• "simple": 2-letter + 6-digit accession numbers are generated. The 2-letter code is defined by annotations$entry_prefix, the 6 digit code is generated from the position of the spectrum in spectraList and the shift given in accessionNumberStart. Leading zeros are added if necessary. (Example: accession numbers XX000043, XX000045 ... will be generated for the first, second ... spectrum in the data-dependent scan if accessionNumberStart is set to 32.) • "selfDefined": Accession numbers are generated by a user-defined function given in accessionBuilderFile. In particular, there is no constraint on the prefix and annotations$entry_prefix will be ignored, if this option is chosen. The function definition must be in the form accessionBuilder <- function(cpd, spectrm, subscan). Note: This functionality is quite advanced. If you really want to specify your own accessionBuilder instead of using the "simple" or "standard" option, we highly encourage you to familiarize yourself with the source code of the function .buildRecord.RmbSpectraSet in buildRecord.R first.
• accessionNumberShifts: A list defining the starting points for generating MassBank record accession numbers. This will be used if accessionBuilderType is unspecified or "standard" (see accessionBuilderType above).
• accessionBuilderFile: A file with a user-defined function to generate MassBank record accession numbers. This will be used if accessionBuilderType is "selfDefined" (see accessionBuilderType above.)
• accessionNumberStart: An integer < 1000000 defining the starting point of MassBank record accession numbers. This will be used if accessionBuilderType is "simple". (see accessionBuilderType above).
• recalibrateBy: Which parameter to use for recalibration: dppm (recalibrate the deviation in ppm) or dmz (recalibrate the m/z deviation).
• recalibrateMS1: Whether to recalibrate MS1 data points separately from MS2 data points ("separate"), with the same recalibration curve as the MS2 data points ("common") or not at all ("none"). Note that the MS1 datapoints points will be used to generate the MS2 recalibration curve in all cases (since this makes the recalibration curve in high-m/z regions better-defined) but may be recalibrated independently themselves, if desired.
• recalibrator: Sets the functions to use for recalibration. Defaults to list(MS1="recalibrate.loess", MS2="recalibrate.loess") which uses a Loess non-parametric fit to generate a recalibration curve. Any custom function may be specified. The function is expected to accept a dataset with variables recalfield and mzFound and to return an object which can be used with predict(). The input recalfield is the value to be estimated by recalibration - it will either contain delta ppm values or absolute mass deviations, depending on the setting for recalibrateBy. In addition to recalibrate.loess, recalibrate.MS1 is predefined, which uses a GAM model for recalibration and appears to work well for pure MS1 datapoints. However, common recalibration for MS1 and MS2 appears to be the best option in general.
• multiplicityFilter: Define the multiplicity filtering level. Default is 2, a value of 1 is off (no filtering) and >2 is harsher filtering.
• titleFormat: The title of MassBank records is a mini-summary of the record, for example "Dinotefuran; LC-ESI-QFT; MS2; CE: 35\%; R=35000; [M+H]+". By default, the first compound name CH$NAME, instrument type AC$INSTRUMENT_TYPE, MS/MS type AC$MASS_SPECTROMETRY: MS_TYPE, collision energy RECORD_TITLE_CE, resolution AC$MASS_SPECTROMETRY: RESOLUTION and precursor MS$FOCUSED_ION: PRECURSOR_TYPE are used. If alternative information is relevant to differentiate acquired spectra, the title should be adjusted. For example, many TOFs do not have a resolution setting. See MassBank documentation for more. • filterSettings: A list of settings that affect the MS/MS processing. • ppmHighMass, ppmLowMass: values for pre-processing, prior to recalibration. The default settings (for e.g. Orbitrap) is 10 ppm for high mass range, 15 ppm for low mass range (defined by massRangeDivision) • massRangeDivision: The m/z value defining the split between ppmHighMass and ppmLowMass above. The default m/z 120 is recommended for Orbitraps. • ppmFine: This defines the ppm cut-off post recalibration. The default value of 5 ppm is recommended for Orbitraps. • prelimCut, prelimCutRatio: Intensity cut-off and cut-off ratio (in \% of the most intense peak) for pre-processing. Affects peak selection for the recalibration only. Careful: the default 1e4 for Orbitrap LTQ positive could remove all peaks for TOF data and will remove too many peaks for Orbitrap LTQ negative mode spectra! • specOKLimit: MS/MS must have at least one peak above this limit present to be processed. • dbeMinLimit: The minimum allowable ring and double bond equivalent (DBE) allowed for assigned formulas. Assumes maximum valences for elements with multiple possible valences. Default is -0.5 (accounting for fragment peaks being ions). • satelliteMzLimit, satelliteIntLimit: Cut-off m/z and intensity values for satellite peak removal. All peaks within the m/z (default 0.5) and intensity ratio (default 0.05 or 5 \%) of the respective peak will be removed. Applicable to Fourier Transform instruments (e.g. Orbitrap). • findMsMsRawSettings: Parameters for adjusting the raw data retrieval. • ppmFine: The ppm error to look for the precursor in the MS1 (parent) spectrum. Default is 10 ppm for Orbitrap. • mzCoarse: The error to search for the precursor specification in the MS2 spectrum. This is often only saved to 2 decimal places and thus inaccurate and may also depend on the isolation window. The default settings (for e.g. Orbitrap) is m/z=0.5 for mzCoarse. • fillPrecursorScan: The default value (FALSE) assumes all necessary precursor information was available in the mzML file. A setting of TRUE tries to fill in the precursor data scan number if it is missing. Only tested on one case-study so far. See also the manpage ?RmbSettings for a description of all RMassBank settings. # The workflow ## Mass spectrometry workflow In the first part of the workflow, spectra are extracted from the files and processed. In the following example, we will process the narcotics spectra from the RMassBankData package. For the workflow to work correctly, a settings file (generated as above and edited accordingly) before must be loaded first. loadRmbSettings("mysettings.ini")  (Note: the template file generated by RmbSettingsTemplate() has no OpenBabel directory specified. Correspondingly, RMassBank will use the CACTUS service instead to generate MOL files. For your actual use, it is strongly recommended to install OpenBabel and specify its install directory in the settings! The CACTUS structures are visually less appealing since they have all hydrogen atoms explicit, and CACTUS is only a backup solution.) First, create a workspace for the msmsWorkflow: w <- newMsmsWorkspace()  The full paths of the files must be loaded into the container in the array files: files <- list.files(system.file("spectra", package="RMassBankData"), ".mzML", full.names = TRUE) basename(files) # To make the workflow faster here, we use only 2 compounds: w@files <- files[1:2]  Note the position of the compound IDs in the filenames. Historically, the "pos" at the end was used to denote the polarity; it is obsolete now, but the ID must be terminated with an underscore. Additionally, the compound list must be loaded using loadList (here, using the formerly copied list from RMassBankData): loadList("./Compoundlist.csv")  This creates a variable compoundList in the global environment, which stores the compound data. Now, we can start the complete workflow to extract [M+H]+ spectral data. The workflow standard workflow consists of 8 steps. The argument archivename specifies the prefix under which to store the analyzed result files. The argument mode specifies the processing mode: pH (positive H) specifies [M+H]+, pNa specifies [M+Na]+, pM specifies [M]+, mH and mFA specify [M-H]- and [M+FA]-, respectively. (I apologize for the naming of pH which has absolutely nothing to do with chemical pH values.) Basically, this runs through the entire workflow, which is explained in more detail below: Step 1: using the function findMsMsHR, all the files in files are searched for MS2 spectra of their respective compound. The found spectra are stored in the array specs. Step 2: A molecular formula fit is attempted for every peak, using the molecular formula of the parent compound as limiting formula, using the function analyzeMsMs. The results are stored in the array analyzedSpecs. Step 3: The analyzed spectra from the array analyzedSpecs are aggregated into the list aggregatedSpecs. This uses the function aggregateSpectra. Step 4: Using the function recalibrateSpectra, a recalibration curve is calculated from the peaks in aggregatedSpecs, and all spectra from specs are recalibrated using this curve. The result is stored in recalibratedSpecs. The recalibration curve is stored in rc. Step 5: The recalibrated spectra (recalibratedSpecs) are re-analyzed with analyzeMsMs and the results stored in analyzedRcSpecs. Step 6: The reanalyzed recalibrated spectra are aggregated with aggregateSpectra into aggregatedRcSpecs. Unmatched peaks in aggregatedRcSpecs are cleaned from known electronic noise using cleanElnoise. A backup copy of all present results is saved as archivename.RData. Step 7: Using reanalyzeFailpeaks, all unmatched peaks from spectra in aggregatedRcSpecs are reanalyzed, allowing$N_2O$as additional elements (to account for oxidation products and$N_2\$ adducts). The results are stored in reanalyzedRcSpecs. A backup copy of all present results is saved as archivename_RA.RData Step 8: The function filterMultiplicity is applied to the peaks: Peaks which occur only once in all analyzed spectra of a compound are eliminated. The filtered list is stored under refilteredSpecs, and a final version of all results is saved as archivename_RF.RData. Additionally, filterMultiplicity creates a CSV file with a list of (relatively) high-intensity unassigned peaks with the name archivename_Failpeaks.csv, which should be manually checked. Peaks to include must be marked with OK = 1.

The steps can be called individually using the steps parameter of msms_workflow. Using the newRecalibration parameter, one can specify if RMassBank should do a new recalibration (default, TRUE) or use the recalibration curve stored in rc (FALSE). This is useful for re-using a recalibration curve in the reanalysis of the same data in another mode: After the detection and processing of all [M+H]+ spectra, which will be present for a large number of compounds, one can rerun the workflow with newRecalibration = F, mode="pNa" and reuse the same calibration curve for Na adduct spectra (which on their own would be too few for a sufficiently good recalibration curve.) The useRtLimit parameter activates or deactivates the usage of retention time constraints when searching for spectra with findMsMsHR.

It is useful to perform the workflow in two blocks, the first being step 1-4 and the second being 5-8. After step 4, a graph is displayed which allows the user to visually evaluate the performance of the recalibration. The top graphs show the distribution of the mass deviation of MS/MS fragments from the predicted mass and the recalibration curve calculated from them; the bottom graphs show the mass deviation of MS precursor ions. The graph to the left is a complete xy plot while the graph to the right is a 2D histogram (if the package gplots is installed on the user's computer).

TODO: Workflow execution in Chunk 10 is currently disabled, I execute Chunk 11 instead for steps that are already done.

w <- msmsWorkflow(w, mode="pH", steps=c(1:4), archivename =
"pH_narcotics")


The recalibration can also be plotted at a later stage:

plotRecalibration(w)


If you are experimenting with new datasets which might give errors, it is advised to run the workflow step by step. This is because if an error occurs, you will lose all intermediate results from the workflow, which might complicate finding the errors. (E.g., if you process steps 2-4 and an error occurs in step 3, you will lose the results from step 2.)

    w <- msmsWorkflow(w, mode="pH", steps=1)
w <- msmsWorkflow(w, mode="pH", steps=2)
w <- msmsWorkflow(w, mode="pH", steps=3)
# etc.


It can be useful to check if any data is retrieved at step 1:

lapply(w@spectra,function(s) s@found)


To check the progress through the workflow, call e.g.:

findProgress(w)


Note that usually a recalibration curve should be done which >15 compounds, and it will become smoother with more compounds. To show the curve found with the full dataset, we can load the preprocessed dataset from the RMassBankData package in another workflow container.

# In the really evaluated workflow, we do the following:
# we run steps 1 through 3, load the recalibration curve from a stored workflow
# and recalibrate the data using that curve.
package="RMassBankData"))


Since this recalibration curve was calculated from a MassBank run of the whole 15 file-dataset, we can copy it into our workspace and use it to recalibrate our data without making a new recalibration curve:

# Just to display the recalibration curve as calculated from
# the complete dataset:
storedW <- msmsWorkflow(storedW, mode="pH", steps=4)
# Copy the recalibration to workspace w and apply it
# (no graph displayed here)
w@rc <- storedW@parent@rc
w@rc.ms1 <- storedW@parent@rc.ms1
w <- msmsWorkflow(w, mode="pH", steps=4, archivename =
"pH_narcotics", newRecalibration = FALSE)


The second part of the workflow can then be processed:

w <- msmsWorkflow(w, mode="pH", steps=c(5:8), archivename =
"pH_narcotics")


If the workflow is performed manually, the results can be stored at any time using

archiveResults(w, filename)


where the former writes the results to a file and the latter duplicates the R objects with a prefix in front of their names. (Note that during the whole workflow, the results are stored automatically after steps 6, 7, and 8 if an archivename is given. So the archivename) parameter is only pro forma for the steps 1-5, but can be added for consistency.

Result files from the workflow on the RMassBankData narcotics spectra dataset are included in RMassBankData, including a marked Failpeaks.csv list.

## MassBank record workflow

An analyzed spectral dataset can then be processed to produce MassBank records. This is done in two major steps: First, annotations for all compounds are retrieved from the Internet, if they are not already present from previously compiled spectra (e.g. if an Internet annotation has already been used to create a [M+H]+ spectrum, it can be reused in the [M-H]- spectrum automatically.)

First, a workspace for the MassBank results must be created starting from processed msmsWorkflow results, and potential pre-existing infolists must be loaded.

To illustrate the workflow, a half-complete annotation list is included in RMassBankData.

mb <- newMbWorkspace(w)
mb <- resetInfolists(mb)
package="RMassBankData"))


Usually, one would call the function with a personal folder:

mb <- resetInfolists(mb)


If we checked the Failpeaks.csv from the previous step and found some important peaks we want to add manually, we can do so and load the peaks into the additional_peaks array:

mb <- addPeaks(mb, my_corrected_Failpeaks.csv)


Now, the record generation workflow can be started:

mb <- mbWorkflow(mb, infolist_path="./Narcotics_infolist.csv")


For all the compounds which were not in the infolists in the infolists_incomplete folder, an entry is fetched and written to Narcotics_infolist.csv (if no infolist_path is specified, the default path is ./infolist.csv.) This file should then be edited and fixed by hand. The entries don't have to be complete; mandatory fields are: at least 1 name, the formula, the exact mass, SMILES code, InChI standard code, InChI standard key. Common errors which must be fixed by hand: 2 near-identical names in the infolist; a very high ChemSpider ID where a lower one exists (which is "better"), a ChEBI entry saying "ChEBI" instead of the actual ChEBI code.

CAUTION: At this stage the compound name is taken from the user-provided compound list and one IUPAC entry from CTS. Please check your compound list carefully! The original naming system from CTS will be reinstated once the scoring system is re-included in the new services.

After fixing the CSV infolist, it should be copied into the infolist folder and the infolist reloaded:

mb <- resetInfolists(mb)


For simplicity / easy testing, a full list for the narcotics dataset is included in RMassBankData:

mb <- resetInfolists(mb)


When we run the workflow again, the line "no new data added" means that the infolists were complete and the workflow can therefore continue.

mb <- mbWorkflow(mb)


The workflow goes through the following steps: Step 1: For compound IDs not in a loaded infolist, new data is fetched from the CTS using the function gather.data and stored in mbdata in tree-like format. Step 2: If new data was retrieved, it is exported to the infolist_path in flat-table format and the workflow stops, otherwise the workflow continues. Step 3: The infolists loaded with loadInfolists are transformed into tree-like MassBank compound information with readMbdata and stored as mbdata_relisted. Step 4: Using the function compileRecords, the compound information from mbdata_relisted is combined with the spectral data and peak lists from aggregatedRcSpecs and refilteredRcSpecs to create compiled records (stored in compiled). All compiled records with at least one good spectrum per compound are in compiled_ok. Step 5: The function toMassbank converts the records into text-file arrays, stored in mbfiles. Step 6: Molfiles are generated for all compounds using createMolfile and stored in molfiles. Step 7: The data stored in the R variables mbfiles and molfiles is written to physical files using exportMassbank in a subfolder named after the MassBank entry prefix. Step 8: A list.tsv file is created using makeMollist.

Subsequently, the two folders moldata and recdata can be zipped and uploaded. This wasn't automated because the Windows version of zip needs additional installed tools.

Note: here, step 6 uses molfile data generated by CACTUS. As stated above, it is strongly recommended to install OpenBabel and add its path to the configuration file for use in mbWorkflow step 6.

# Session information

sessionInfo()


## Try the RMassBank package in your browser

Any scripts or data that you put into this service are public.

RMassBank documentation built on Nov. 8, 2020, 6:06 p.m.