4 Mass Data Integrations Featured by MineXpert

Each time an integration is performed, MineXpert stores processing information data that permits to characterize the integration for later internal reuse. Any integration, as it is performed, documents itself to the Processing Flow. For example, one integration might be from TIC chromatogram to mass spectrum, another integration can be from mass spectrum to drift spectrum. These kinds of integration are documented as Processing Types . For each integration, there is thus a source and a destination. But this is not sufficient to document precisely the integration: there must be other informational data along with the integration type. When integrating from TIC chromatogram to mass spectrum, the user might select only a region of interest in the chromatogram. That is another kind of information that is documented: the source range (in this case a retention time range). Likewise, when integrating from a mass spectrum to a drift spectrum, the origin is a m/z range and the destination is a drift spectrum. Other data are stored in the processing information data, like the fragmentation specification—if any has been set— and the m/z integration parameters, in case the integration produces a mass spectrum.

Together, all the processing information data are stored in what are called Processing Flow instances. A processing flow instance can contain a default MSⁿ fragmentation specification and/or default m/z integration parameters. It can contain any number of Processing Step instances that in turn can contain any number of Processing Spec instances. Like for the processing flow, each processing step can contain a MSⁿ fragmentation specification and/or m/z integration parameters.

How is the processing flow documented ? When a mass spectrometry data file is loaded from disk, MineXpert iterates in all the mass spectra of the MS run data set and computes a TIC chromatogram. That TIC chromatogram is displayed in the TIC/XIC chromatograms window. At this point, no MSⁿ fragmentation specification is set and no m/z integration parameters are set. When the user starts exploring the data, their first integration will determine how the processing flow will be filled-in with details about that integration.

If an integration is performed from the TIC chromatogram to a mass spectrum, then m/z integration parameters will necessarily be used. Either the user has configured these parameters (Section 4.1.3, “ Setting the m/z Integration Parameters ” ) and they will be used to perform the mass spectral combinations, or default values will be used (these default values are crafted from the statistical analysis of the whole MS run data set). The processing step that is created to document this specific integration will thus document the m/z integration parameters that have been used for the mass spectra combination. That processing step is added to the processing flow that documents the whole integration process. The plot that is created in the Mass spectra window will have, associated to it, that processing flow. Think of the Processing Flow as a pedigree ID card that each plot has associated to it. That pedigree gets incremented with new processing steps each time an integration is chained to the previous integrations.

Tip

The original feature of the Processing Flow concept is that, because each plot has its own processing flow, when a new integration is performed starting from that plot, then all the preceding steps of the processing flow are replayed and the new step is added to the processing flow. This has the beneficial effect of virtually restricting the scope of the initial data set into a much smaller data set all along the various integrations that occur during a data exploration session. When the scope reduces, the computing times reduce accordingly. However, the initial data set is not reduced in memory, only the scope of it is reduced. This is much different from other software in which mass data are effectively pruned off the initial data set with the drawback that data must be reloaded each time an integration must start from the initial data set.

At any moment is it possible to set the MSⁿ integration parameters by selecting the Open MS fragmentation dialog menu item from the plot widget main menu (Section 3.7.3, “The Plot Widget Main Menu” ).

The MSⁿ fragmentation parameters might be set at any time from the plot widget's main menu. The criteria are of three kinds, as described in detail in the text.

Figure 4.1: Setting the MSⁿ fragmentation parameters #

The dialog window that is displayed is depicted in Figure 4.1, “ Setting the MSⁿ fragmentation parameters ” . The available settings are explained below:

Note

Note that in the dialog window above, none of the three criteria is essential. Setting the MS level to 0 disactivates that parameter as a valid criterion for filtering spectra during the integration. In this case, only MSⁿ fragmentation specifications used in preceding integrations are taken into account (that is, fragmentation specifications present in the Processing Flow). If one wants to reset the MSⁿ fragmentation specification to a lower MS level than that of the last preceding integration, the only way to achieve this is to go back to the previous plot where the processing flow contained processing steps of the right MSⁿ fragmentation level values. Indeed, it is not possible to reset the MSⁿ fragmentation level in a plot by setting the MS level to 0 because that plot already has a processing flow containing pocessing steps configured with a non modifiable MS level.

Tip

There is a shortcut to the MSⁿ fragmentation MS level in each composite plot widget, as shown below, where the spin box widget on top of the right hand side column of allows one to set the MS level.

If the only fragmentation setting to be modified is the MS level, then, instead of opening the dialog window described above, it is possible to set the required MS level right into the spin box widget at the top of the right hand side column of buttons in the composite plot widget.

Figure 4.2: Shortcut to the fragmentation MS level setting #

The m/z integration parameters are only compulsorily used in case an integration will produce a mass spectrum or a color map involving m/z data in any one of the two dimensions. Indeed, these parameters are required to configure the way mass spectra are combined together into a single product mass spectrum.

An integration to a mass spectrum occurs when the user -selects a range in a given plot while the checkbutton is checked. Integrations to a mass spectrum can be elicited from any plot.

Very rarely can one perform an integration to a mass spectrum without having to first configure the way the m/z values are binned into the combination spectrum. The determination of the best m/z integration parameters is a process that might be required each time a new data set comes from an instrument from which to data have been obtained before. Indeed, there is a vast number of different mass spectral data acquisition/storage modes that depend on the vendors or on the mass spectrometer models. In general, this diversity of mass spectral data files ^[10] creates difficulties in the mass spectra integrations, as shown below in Figure 4.3, “ Unusable combination spectrum without binning ” .

The mass data used to compute this combination spectrum originate from a Lumos Orbitrap analyzer. The visible signal should have been three peaks belonging to a 2-charged ion isotopic cluster. Each peak of the isotopic cluster is artifactually made of numerous data points because the data from the Orbitrap analyzer have 10 decimal digits. Without binning, the mass spectrum is almost unusable.

Figure 4.3: Unusable combination spectrum without binning #

In this combination mass spectrum, computed without binning from Lumos Orbitrap-originating data, three peaks that belong to an isotopic cluster appear as made of a number of “sub-peaks”. This is due to the fact that the number of decimals for the m/z values in the data file is so high that a single isotopic peak appears as a set of peaks. The signal in this mass spectrum is totally useless. This is a perfect illustration of the necessity of data binning of the m/z values of the mass spectra to be combined.

Integrations to any plot that involves m/z as one of its dimension (be that a mass spectrum or a color map) can be configured to ensure the best results, depending on the kind of mass data. Proper binning configuration is key to getting best results.

Figure 4.4: The m/z integration parameters window #

MineXpert provides a number of ways to configure mass spectral combinations such that the obtained mass spectrum is usable. The m/z integration parameters that might be set are described in the following sections. The window depicted in Figure 4.4, “ The m/z integration parameters window ” can be displayed from any plot widget menu using the Open the m/z integration params dialog menu item (see Section 3.7.3, “The Plot Widget Main Menu” ). The following parameters can be set:

When all the parameters have been set, click onto Apply or key in CTRL RETURN . At this point the next integrations started from the plot widget in which these m/z integration parameters were set will be performed using the settings. These parameters will propagate to all the descendant widgets. It will be possible, at any moment and in any widget, to modify the parameters again.

Tip

If the check box \guilabel{Set as default} is checked, the system records these parameters and makes them the default ones for any coming integration.

This section provides guidelines on the best way to determine the m/z integration parameters that will result in nicely shaped mass peaks in the combination mass spectrum. There might be situations where there is no need to set the m/z integration parameters, typically if the mass data have been properly binned by the software running the mass spectral data acquisition (Waters and Agilent systems often provide nicely binned mass spectra).

However, in a vast majority of the cases, proper setting of the m/z integration parameters will be essential to achieve a combination of mass spectra resulting in a useful mass spectrum. In particular, it is important to grasp that the binning might need to be dynamically set while combining (m/z,i) pairs along the m/z axis. This dynamic binning is configured by choosing either res or ppm (for resolving power or part-per-million, respectively; bins have varying sizes throughout of the whole m/z range).

When dealing with a mass spectrometry data set from an instrument that is unknown, the first step is to perform an integration to a mass spectrum without binning. In this situation, all the distinct peaks observed in all the mass spectra being combined together are going to be stored in the combination spectrum. When the data come from an instrument that does not produce nicely binned mass spectra, the resulting combination mass spectrum might be highly spiky, with spurious data points interspersed between seemingly correct data points. This is typically visible in the mass spectrum shown in Figure 4.5, “ Determination of the best m/z integration parameters for integration to mass spectra ” .

Zoomed in view of a protein mass spectrum.

Figure 4.5: Determination of the best m/z integration parameters for integration to mass spectra #

The different plots in the figure show the mass spectral data combined for the same acquisition data and for the same retention time range. From top to bottom:

Dragging the mouse cursor between the extremities of an isotopic cluster peak, at half of its height, provides a rough estimation of the resolving power of the instrument.

Figure 4.6: Estimating the resolving power of a mass spectral acquisition #

One evidence that the setting for this integration is correct is that it is now possible to use the inter-peak deconvolution method for the determination of the ion charge (36+), which confirms the result of a previous deconvolution based on the charge state envelope.

Figure 4.7: Determination of the ion charge by using the isotopic cluster #

One important concept to grasp is that the bin size used for the mass spectra combinations will actually determine the final resolution of the mass spectrum (or color map). For example, imagine that the setting is the following: Arbitrary binning with bin size of 1~Da; in that case the mass spectrum or the m/z-based color map will be very rough. Instead, setting the bin size to 0.05 will provide a much better spectrum (or color map). Of course, going down with the size of the bins will increase almost linearly the computation time required to go through them all while adding signal from the various spectra that are being combined. This is exemplified in Figure 4.8, “ Effects of the bin size on the color map involving one m/z dimension ” .

The three different color map regions were computed in the Arbitrary binning mode with bin sizes 1~Da, 0.1~Da and 0.01~Da (top to bottom).

Figure 4.8: Effects of the bin size on the color map involving one m/z dimension #

Tip

Because the computation time increases inversely with the size of the bins, it is recommended to start a color map-based data exploration session with an integration over the whole data set but with large bins (typically 1~Da) that will provide enough sight on the data. Then, when the feature of interest has been located on the color map to a given region, reduce the bin size to increase the resolution of the color map and integrate from that same region to a new color map.

Mass spectrum data points with an intensity value equal to 0 might arise either from the initial data, as read from the file, or from the binning process, as detailed below.

When bins are prepared, prior to starting a mass spectral combination operation, they all correspond to a m/z value associated to a 0-value intensity. Once the bins have been prepared, the combination starts and, while iterating through all the data points of all the spectra, fills-in the bins. If, however, a bin is never filled (it ends up never updated during the combination), that bin will have an intensity of 0. Bins having a 0-intensity value have a bad effect on the result mass spectrum. Removing them by setting the Remove 0-value data points parameter proves beneficial, as shown in Figure 4.9, “ Removing 0-intensity data points ” . When the 0-intensity data points are not removed (upper spectrum), the signal is deteriorated by spurious inverted spikes. Removal of the 0-intensity data points, cleans the trace perfectly.

When arbitrary binning is performed, residual 0-intensity data points might survive in the combination spectrum, which deteriorates the resulting mass spectrum. Removing these data points from the combined mass spectrum cleans the trace.

Figure 4.9: Removing 0-intensity data points #

The Savitzky-Golay filtering method is widely known for its effectiveness in removing noise from mass spectral data. It is possible to apply that filter to any plot by selecting the corresponding menus as described at Figure 4.10, “Menus to configure and run the Savitzky-Golay filter to smooth traces ” . The Savitzky-Golay parameters can be set by selecting the Configure Savitzky-Golay smoothing filter menu as shown on Figure 4.10, “Menus to configure and run the Savitzky-Golay filter to smooth traces ” .

The Savitzky-Golay algorithm is a powerful filter that allows one to smooth any graph. The first menu enables the configuration of the filter and the second menu actually applies the filter to the trace.

Figure 4.10: Menus to configure and run the Savitzky-Golay filter to smooth traces #

The Savitzky-Golay filter requires careful configuration. The configuration provided by default is shown in Figure 4.11, “ Configuration of the Savitzky-Golay filter to smooth traces ” .

The different parameters used to configure the Savitzky-Golay algorithm are described in the text.

Figure 4.11: Configuration of the Savitzky-Golay filter to smooth traces #

The centroidation of a mass spectrum is also called peak picking. This process is aimed at converting a profile-mode spectrum, that has its peaks properly shaped in Gaussian-like form, into a set of m/z values that recapitulate the initial peaks.

One aspect that is very important in peak picking is the determination of what is considered to be noise and what is considered to be actual signal. The signal-to-noise estimation is an optional task that is only carried out when the Signal to noise ratio is set to a value greater than 0. The signal-to-noise estimator needs a number of parameters to be set up (in the Configuration of the signal to noise estimator group box widget). One important configuration bit is related to the determination of what intensity value is considered to be the maximum intensity value over the whole mass spectrum. There are multiple ways to define that maximum intensity value, as described in ??? .

Other configuration bits are the Rolling window size which specifies the m/z range considered when computing the local signal-to-noise ratio. The rolling window is moved data point by data point over the whole mass spectrum. Each mass spectrum data point has thus a corresponding signal-to-noise ratio value. The Mininum required elements configuration bit is the number of data points that are required to be found inside the rolling window. The Bin count configuration bit is the number of bins in the histogram of the intensities in the mass spectrum. This parameter defines the granularity with which the signal-to-noise ratio is computed for each mass spectrum data point.

Figure 4.12: Configuration of the centroidation of mass spectra #

Finally, one last bit of configuration for the centroid extraction is the Allowed missing peaks value that defines the number of m/z values that can be found to be absent from the rolling window. This parameters provides some leeway when dealing with sparse mass spectra.

When the centroid extraction process is terminated, the centroids are plotted as a bar graph. These centroid data points can be exported to a file or a clipboard exactly the same way as for the normal trace graphs.

Loading data from mass data files in mzML format does not guarantee that the data will be of the same kind when they originate from different mass spectrometers. For example, data from Orbitrap mass spectrometers have the following characteristics:

This is the output of the statistical analysis of the data loaded from a Lumos Orbitrap-originating file:

As mentioned earlier, the most interesting bit of information is in the line reproduced below:

That 0.0062 value somehow gives an indication of the “definition” of the spectrum, that is, the smallest distance possible between two consecutive points in the m/z-axis.

In general, the fact that the spectra of an acquisition do not all have the same m/z vector as the m/z-axis is a great difficulty for mass spectral integration because it requires setting up binning prior to performing the mass spectral combination. That binning is nothing else than crafting a m/z value vector able to receive the intensities of all the m/z data points in the spectra to be combined. These concepts are developed in the next paragraph.

At the end of the data file loading, MineXpert performs a rudimentary statistical analysis of the data. The main datum of interest is the smallest m/z step that is observed in the whole set of mass data loaded from disk (the mass spectrum list, that can hold mass spectra in the thousands). For each mass spectrum in the list, the smallest m/z delta between any two consecutive data points is recorded. Then, the smallest ever m/z delta value is sought amidst all the recorded values. Intuitively, that smallest m/z delta value provides an idea of the resolution power of the instrument that generated the mass spectra. Interestingly, this is not the proper value to configure binning. The best value is the median value of the smallest m/z delta values encountered over all the mass spectra of the data file. It is the value that is suggested by default to arbitrarily construct the bins during an integration to a mass spectrum, as described in Figure 4.4, “ The m/z integration parameters window ” (Arbitrary binning value with bin size unit dalton).