Before diving into the structure of mass spectrometry data, processing work-flows, statistical tools, and knowledge representations, we need to have a clear idea of the purpose of a project; in the absence of a well-defined question, any further effort is pointless.

The human mind tends to build empirical knowledge from individual observations and experiences. Knowledge obtained this way might be of practical use, but does not withstand rigorous scientific criteria. Sir Karl Raimund Popper (1959) highlighted such thinking errors by the use of a simple example:

Now it is far from obvious, from a logical point of view, that we are justified in inferring universal statements from singular ones, no matter how numerous; for any conclusion drawn in this way may always turn out to be false: no matter how many instances of white swans we may have observed, this does not justify the conclusion that all swans are white.¹ 

This problem of induction was investigated by David Hume (1748) who questioned the causality of matters of fact which are not supported by deductive logic.² 

According to The Scientific Method of Popper, the creation of universally valid models requires a hypothesis as a starting point. Deductions and statements are driven by the hypothesis, which, in turn, can be tested theoretically or experimentally. Popper introduced the concept of falsifiability. Often, hypotheses or theories cannot be verified by experimental evidence, but individual deduced statements can be falsified by testing to evaluate their limits.¹  The differences between deductive and inductive approaches are shown in Figure 1.1.

Returning from epistemology to practical experimental design, two main strategies are distinguishable, namely hypothesis-driven and the exploratory approaches. Having a well-defined hypothesis-driven analytical question greatly facilitates experimental planning. For example, if the regulation of a certain biosynthetic pathway is proposed, related intermediates and products can be monitored for with a targeted metabolomics strategy, which increases the quality of the data for the compounds of interest and reduces the acquisition of less relevant (for this question) data. In addition, alternative analysis methods, such as HPLC, optical methods, ELISA testing, and complementary studies, such as qPCR, are considered to increase the sample number and further strengthen the necessary evidence.

In stark contrast, many omics studies lack a clearly defined hypothesis; rather, they strive to explore possible differences between two or more groups of organisms on a particular level, for example a transcriptome, proteome, or metabolome. Therefore, the expected outcome of comparative omics experiments are patterns that support the formulation of hypotheses.⁴ 

Data mining (DM) is considered to be a hybrid between deductive and inductive research. Although DM is data driven and lacks human creativity, predictive models can be built within a defined numerical space. DM extends ‘classic’ statistical tools such as principal components analysis (PCA) by artificial intelligence (AI) and machine learning. The outcomes of a DM procedure are classification models, associations, and the importance of variables. An excellent practical guide to DM using the R statistical software was published by Williams (2011).⁵ 

The training of a new model starts with a question that should be answered using available data. In case of a metabolomics study, possible questions might include:

• Do the wild types of a genotype and its mutant display distinct metabolic identities? ⇒ Classification model.

• If so, which metabolites are indicative for each group? ⇒ Variable importance analysis.

• Which level of glucose can I expect for parameter set X (which is not part of the training data)? ⇒ Quantitative model.

• Which of the metabolites directly or indirectly interact with each other? ⇒ Association analysis.

Importantly, only part of the dataset is used to construct the actual model; two other parts of the data set serve for validating the model and testing its performance, for example, to estimate a realistic classification error.⁶ 

For example, the random forest (RF) algorithm deals well with mass-spectrometry-based omics data that are characterized by a large number of variables and relatively few observations. Furthermore, RF analysis is tolerant to noisy data and may even be employed as a filter to discriminate between informative signals and random ones.

It is worth noting that DM methods are not knowledge biased and do not give meaning to signals other than statistical ones; therefore, they can help the scientist to discover unexpected relationships in a dataset. Altogether, DM methods are very useful for the processing of MS data, and their practical value (e.g., for classification models in biology and medicine) is only now becoming recognized.⁷ 

John Bennet Fenn, who shared the Nobel Prize in Chemistry (2002) with Koichi Tanaka and Kurt Wüthrich for the development of methods for identification and structure analyses of biological macromolecules (https://www.nobelprize.org/prizes/chemistry/2002/summary/), described the concept of mass spectrometry in a few words:

Mass spectrometry is the art of measuring atoms and molecules to determine their molecular weight. Such mass or weight information is sometimes sufficient, frequently necessary, and always useful in determining the identity of a species. To practice this art one puts charge on the molecules of interest, i.e., the analyte, then measures how the trajectories of the resulting ions respond in vacuum to various combinations of electric and magnetic fields. Clearly, the sine qua non of such a method is the conversion of neutral analyte molecules into ions. For small and simple species the ionisation is readily carried by gas-phase encounters between the neutral molecules and electrons, photons, or other ions. In recent years, the efforts of many investigators have led to new techniques for producing ions of species too large and complex to be vaporized without substantial, even catastrophic, decomposition.

Many different solutions that address the technical challenges of mass spectrometry have been found, and only basic MS principles are expounded upon in this section, which should provide the reader with sufficient knowledge to understand data provided by mass spectrometry. A more detailed introduction can be found in the book The Expanding Role of Mass Spectrometry in Biotechnology by Gary Siuzdak (2006),⁸  for example.

The acronyms used in this book follow the glossary of the Analytical Methods Committee (AMC) of the Royal Society of Chemistry;⁹  other terms are used in accordance with IUPAC Recommendations 2013.¹⁰ 

As illustrated in Figure 1.2, a mass spectrometric system consists of the components described below.

Although this might appear to be trivial, the sample needs to be transferred into the mass spectrometer. Whereas the sample is at ambient temperature and pressure, the mass analyser requires vacuum conditions; hence, the sample stream entering the mass analyser must be free of liquids (‘dry’) and excess gas needs to be removed by differential pumping. For calibration or the direct flow injection (DLI) of a small number of individual samples, solutions can be introduced using a syringe pump, while an autosampler is usually used for larger sample numbers. To avoid ‘batch effects’ during data acquisition and analysis, it is strongly recommended that samples be analysed in random order.

Fractionation of a complex sample improves the detection and quantification of compounds, because matrix effects and ion suppression are reduced. Consequently, mass spectrometry is often augmented with gas or liquid chromatography (GC and LC). The retention times of the analysed compounds provide additional analytical dimensions for identification purposes. During mass spectrometry imaging (MSI), surfaces are scanned and mass spectra are combined with the location data of each sampled spot. MSI generates images that visualize the distribution of mass signals in a planar projection.

To activate neutral molecules for mass spectrometry, they need to be converted into positive or negative ions. This process occurs in the ion source. A neutral molecule ‘M’ can be charged by different processes:⁸ 

Positively charged ions are formed by:

• Protonation M + H⁺ → MH⁺

• Cationization M + cation⁺ → [M + cation]⁺

• Electron ejection M − e⁻ → M⁺.

Negatively charged ions can be generated by:

• Deprotonation M − H⁺ → [M − H]⁻

• Electron capture M + e⁻ → M⁻.

Ion sources are classified as ‘soft’ or ‘hard’, depending on the ionization energy, with the latter providing a higher degree of fragmentation. The type of ionization defines which types of molecular ion are expected to be observable. The most frequently used types of ion sources in metabolomics and proteomics are listed in Table 1.1.

Electron ionization (EI) is the method of choice for highly volatile molecules with low polarity (‘hydrophobic’ compounds), e.g. in drug metabolomics, lipidomics and forensics.¹³  Due to the high energy level of 70 eV,⁸  organic molecules are extensively fragmented and the intact M⁺˙ signal is generally of low intensity or absent.¹²  Usually, EI MS is coupled to gas chromatography (GC). GC-EI MS is highly standardized and has been employed in biochemical labs for decades. Therefore, numerous free and commercial databases with reference spectra are available and facilitate the identification of compounds.^13,14 

Hydrophobic gas phase molecules also can be ionized by chemical ionization (CI). Compared to EI, the energy level is lower. Therefore, the fragmentation is reduced and the detection of molecular ions with sufficient intensity is more likely.⁸ 

Atmospheric pressure chemical ionization (APCI) and atmospheric pressure photoionization (APPI) ionize vaporized neutral molecules with medium and low polarity. Compared to APCI and ESI, APPI is more robust regarding buffer salts and ion suppression,¹⁵  and ionizes more hydrophobic compounds such as polycyclic aromatic hydrocarbons (PAHs).¹⁶ 

Electrospray ionization (ESI) favours the creation of protonated molecules and positively charged adducts in positive mode, and deprotonated molecules in negative mode. The samples enter the ion source with a solvent flow, either by direct liquid introduction (DLI), or from a coupled LC. Mainly polar compounds are ionized, including many organic small molecules, lipids (except for nonpolar lipids, e.g. sterols and triacylglycerols), peptides and proteins. ESI is a very soft ionization method with low levels of fragmentation, which allows the detection of intact biomolecule ions. Even complete viruses can be observed by ESI while maintaining their virulence.¹⁷  Large biomolecules present multiple charges ([M + nH]ⁿ⁺), resulting in ions of a mass-to-charge ratio (m/z) that can be analyzed with most mass analyzers (see below, Table 1.2).¹⁸  ESI with very low flow rates, nanoelectrospray ionization (nanoESI), is especially useful in proteomics, because of its increased tolerance against buffer salts and high sensitivity.¹⁹  Since biological samples often contain alkali metals (i.e., sodium and potassium), [M + Na]⁺ and [M + K]⁺ ions are often observed in ESI mass spectra. The addition of ammonium acetate can be used experimentally to dissociate alkali metal adducts, leading to the formation of protonated or [M + NH₄]⁺ ions instead. In contrast, the addition of lithium can be used to ionize sugars by cationization.²⁰  ESI/nanoESI the most commonly used ion source in LC-MS metabolomics and proteomics, because of its wide m/z range and huge diversity of detectable organic molecules.

Matrix-assisted laser desorption/ionization (MALDI) is a soft method for large biomolecules.²¹  MALDI requires the co-crystallization of the analyte molecules with a matrix. Depending on the choice of matrix, many different compound classes can be ionized (e.g. proteins, lipids, DNA). Since the prepared samples need to be solid, a direct coupling of MALDI to LC-MS systems is not practical. In contrast, MALDI is frequently used in mass spectrometry imaging (MSI)²²  and high-throughput screening. The main challenges of MALDI are related to the interference of matrix substances with target molecules, and the lacking reproducibility between experiments. Advanced data analysis methods²³  and novel matrix compounds²⁴  therefore can enhance the utility of MALDI.

Mass analyzers require the analyte in a gaseous state (see Figure 1.2). Therefore, molecules that are already charged in solution, still need to be transferred into the gas phase:

This vaporization occurs in various ion sources such as APCI, ESI/nanoESI, MALDI.⁸ 

Mass analyzers separate ions based on their mass-to-charge (m/z) ratios and different types of mass analyzer have inherent differences in their analytical performance. The most advertized parameter of a mass analyzer is its specified mass resolving power or resolution. The value of this parameter indicates to which degree signals of similar mass-to-charge ratio can be distinguished (see Figure 1.3). Resolution is calculated by:

The mass accuracy expresses the deviation of the measured mass m_m from its exact (theoretical) value m_e and can be expressed in terms of m/z tolerance (e.g. ± 0.15 m/z) or relatively, in parts per million (ppm):

Mass accuracy depends on the operational conditions of the MS and its calibration. In addition to external calibration, mass accuracy can be increased by the use of internal calibrants. For example, high mass accuracy is crucial for determining the chemical sum formula of a compound based on its mass-to-charge ratio.

The dynamic range of the mass analyzer is also important for quantitative experiments; this parameter provides the ‘ratio between the largest and the smallest detectable signals’ (http://mass-spec.lsu.edu/msterms/index.php/Dynamic_range). A high dynamic range facilitates the evaluation of minor constituents in complex mixtures containing multiple molecules.

Another important performance parameter of a mass analyzer is its scan cycle time; i.e., the time required to collect a single mass spectrum. For time-critical applications, such as fast metabolomics screening, a mass analyzer or method with low resolution and/or a reduced mass range, among others, is required; however, a shorter scan cycle time might be a preferable option.

Typical specifications of mass analyzers commonly used in metabolomics and proteomics are given in Table 1.2.

Overall, no ideal mass analyzer exists; hence, a compromise is required, which depends on the principle use of the mass analyzer and the available budget.

A triple quadrupole (QqQ) mass spectrometer is an excellent choice for the routine analysis of known compounds, because it provides high sensitivity over a wide dynamic range. QqQs are often used for quantification with multiple reaction monitoring (MRM) methods (see below), e.g. in food trace analysis. The resolution and the mass accuracy of quadrupole analyzers are modest. On the other hand, quadruples are reliable and economic. Therefore, quadrupoles are often used in hybrid MS systems, e.g. as upstream mass filters, and in collision cells.

Ion traps are constructed in two basic models: quadrupole (cubic) ion traps (QITs) and linear ion traps (LITs). In contrast to quadrupoles, that operate with a continuous ion flow, ion traps are able to collect ions. Thus, a low analyte concentration can be compensated by an increased ion collection time. Consequently, ion trap analyzers can reach high sensitivities. The resolution and the mass accuracy of ITs are limited, but the fragmentation of ions is possible to obtain structural information. Thus, LITs can be used for proteomics and metabolomics, if reference data are available (‘spectral matching’). The sequential (manual or automated) selection and fragmentation of ions (‘MSⁿ capability’) allows for in-depth molecule structure studies.

Time-of-flight (ToF) mass analyzers provide high resolution and mass accuracy. At the same time, ToF mass analyzers provide an extended dynamic range, which makes them suitable for chemical profiling/comparative metabolomics studies. Qq-ToF devices additionally provide MS² capability and are widely used in metabolomics and proteomics.

The introduction of the Orbitrap mass analyzers about two decades ago²⁶  was game-changing, because devices with extremely high resolution became more accessible for routine laboratories. Orbitraps are suitable for most metabolomic and proteomic studies, including the analysis of complex mixtures. The high resolution and mass accuracy facilitate chemical structure elucidation and de novo peptide sequencing. Additionally, Orbitrabs can be used for fragmentation studies of entire proteins (‘top-down proteomics’).²⁷ 

Fourier transform ion cyclotron resonance (FT ICR) devices represent the high-end mass analyzers and require significant investments in scientific infrastructure. Since the Orbitrap analyzers are catching up in analytical performance, FT ICR mass analyzers are reserved for the most demanding analytical tasks such as the characterization of proteoforms²⁸  or crude samples of terrestrial dissolved organic matter (DOM).²⁹ 

Ion mobility separation (IMS) determines the ion collision cross section (CCS) of a molecule, which gives information about its geometrical shape.^30,31  This additional analytical dimension permits isobaric compounds (molecules of the same mass) to be discriminated, and the study of (bio)molecular confirmation. IMS forces charged molecules to travel in an electrostatic field against the resistance of atmospheric molecules or a gas steam. An IMS cell operates at an elevated pressure compared to the mass analyzer and, therefore, is usually coupled upstream with the mass analyzer (e.g. LC-IMS-Qq-ToF). Currently, eight different IMS instrument types have been described,³²  which are coupled to mass analyzers, with drift tubes and travelling wave separators being the most common ones.³³  An increasing number of ion mobility-mass spectrometry (IM-MS) devices are now commercially available, which has led to a growing number of reported applications and reference data, which has stimulated the active development of IM-MS software.

The analytical performance of a device depends strongly on its operating conditions, which include calibration, maintenance, installation site (vibrations, electricity, temperature, and humidity), and the device method. For example, ToF analyzers have wide mass ranges that can extend beyond 100 000 m/z, which facilitates the analysis of complete, singly charged proteins. However, sensitivity and resolution will be much below the technical specifications in such an experimental set-up.

To interpret data, practical values of instrumental parameters need to be determined. For example, a mass accuracy of 3–10 ppm is realistic when analysing complex mixtures of metabolites by ToF without internal calibration, compared to the specified value of 1 ppm under optimal conditions. Of course, high mass accuracy facilitates the identification of compounds by database matching and the de novo creation of sum formulas; however, even mass accuracies below 1 ppm are insufficient to identify metabolites by mass alone.³⁴  Taking into account isotope patterns and chemical rules helps to narrow down possible structure candidates, Kind and Fiehn reported ‘Seven Golden Rules’ for the heuristic filtering of chemical formulas.³⁵ 

A combination of different analytical dimensions (MSⁿ analysis, isotope pattern, IMS collision cross section and chromatographic retention time) and/or the co-analysis of an authentic reference material is usually necessary to reliably identify a compound by MS.³⁶  Claiming novel organic structures usually also requires the reporting of X-ray or nuclear magnetic resonance (NMR) data.

A compound ion can be selected and fragmented on a mass analyzer engendered with that capability in order to obtain structural information. With the exception of simple (i.e., without another analyzer coupled) quadruple and ToF MS analyzers, all MS systems have this feature. Depending on the construction of the mass analyzer, fragmentation is either performed in a separate fragmentation cell or in the same cell. Collision-induced dissociation (CID) is provoked if an inert gas is introduced into a chamber containing ions at high vacuum.³⁷  High-energy collision-induced dissociation (HCID) is based on the same principle, but it provides high MS/MS resolution on Orbitrap instruments;³⁸  (H)CID fragmentation is widely used in metabolomics and proteomics applications.

Electron capture dissociation (ECD), electron transfer dissociation (ETD), and other ECD-derived methods that are referred to as ‘ExD’ expose multiply charged ‘parent ions’ to free electrons. The capture of these electrons by these molecules results in their fragmentation. ExD methods provide complementary fragmentation information and are mainly used in MS-based protein studies.³⁹ 

Often, fragmentation spectra are referred to as ‘MS²’ or ‘MS/MS’ spectra, suggesting that they are obtained in a second experiment following the acquisition of the MS scan. However, fragmentation can also occur in the ion source (‘in-source fragmentation’, ‘in-source decay’, or ‘in source collision-induced dissociation’, InS), depending on the stability of the structures and the applied ionization energy. For example, while EI ionization leads to a high proportion of fragmented ions, APCI and ‘soft’ ionization methods also display fragmentation events, for example, by laser-induced photodissociation (PD) during MALDI.

Information about the expected fragmentation behaviour of a molecule greatly supports its structural elucidation or identification.^36,40  Currently, databases of small molecules mainly contain fragmentation spectra following EI ionization and CID, and these reference databases sometimes contain spectra that were collected with different fragmentation energies. Consequently, it might be necessary to collect spectra with different fragmentation settings in order to identify unknown ions.

In proteomics, theoretical peptide fragments are calculated from the amino acid sequence. Depending on where the peptide bond is broken, N-terminal fragment ions are referred to as ‘a, b,’ and ‘c’, while C-terminal ions are labelled ‘x, y,’ and ‘z’; b and y ions are expected with CID or HCID, which are the most common fragmentation methods in proteomics, while c and z ions are mainly expected for ECD/ETD.

The selection and further fragmentation of selected fragment ions is possible in MS systems, resulting in MS³, or even MS⁴ and MS⁵ (etc.) if this process is continued. Such ‘MSⁿ’ fragmentations provide detailed structural information, but require a rather manual procedure with few reference spectra available in databases. Nevertheless, MSⁿ provides valuable structural information from devices with low or modest mass resolutions, such as ion-trap analyzers.

The detector is usually not deliberately chosen by the user, but is an integral and fixed component of an MS system. Consequently, not much attention is usually paid to this critical component. This section is based on information from a concise review of the history of ion current detectors.⁴¹ 

Most mass analyzers require a detector to sense the intensity of ions in a certain m/z region and to translate them into an electrical/computational signal. One ion per second corresponds to 1.6 × 10⁻¹⁹ A, which can technically be measured; however, the minimum detectable ion intensity of a detector is usually about a thousand times higher. The most frequently used detectors are secondary electron multipliers (SEMs), such as micro-channel plate (MCP) detectors. However, Faraday cup, ion-to-photon, and cryogenic detectors have also been used, but the latter ones are mainly employed for special applications, such as the detection of macroions. Instead of a separate ion detector, FT-ICR, and Orbitrap analyzers record the current induced by resonating ions. This process is non-destructive and leads to high analyzer resolution, accuracy, and sensitivity.⁴¹ 

Mass analyzers determine the mass-to-charge (m/z) ratios of introduced ions. Figure 1.3A shows the theoretical mass spectrum of pure protonated nicotine with its sum formula. Based on its chemical composition, its monoisotopic mass can be calculated to be 163.123 m/z. This mass corresponds to the ion that is composed only of the most abundant elemental isotopes. Subsequent less-intense peaks represent ions that contain ₁₃C, ²H, or other isotopes. The pattern of m/z values and their relative intensities for a particular compound is referred to as the ‘isotopic distribution’. The isotopic distribution of protonated nicotine shown in Figure 1.3A was simulated with the R package (http://orgmassspec.github.io/).^42,43  The program uses the data tables for Atomic Weights and Isotopic Compositions with Relative Atomic Masses, maintained by the National Institute of Standards and Technology (NIST) (http://physics.nist.gov/PhysRefData/Compositions/). These data are based on those published by the Commission on Isotopic Abundances and Atomic Weights (CIAAW) of the International Union of Pure and Applied Chemistry (IUPAC) (http://www.ciaaw.org/atomic-weights.htm),⁴⁴  and the new evaluation of atomic masses, Ame2012.⁴⁵ 

Two different m/z-signal representations are shown, namely centroid spectra that consist only of m/z values and their respective intensities, whereas in profile spectra, peaks are formed by curves defined by various data points. Centroid data are either derived from profile spectra (by peak picking) or collected directly during MS in centroid acquisition mode. Since centroid spectra contain less information, their data files are smaller and allow faster scan-cycle times. Most MS data processing workflows convert profile spectra into centroid spectra. On the other hand, profile spectra contain important information about the quality of the MS data and the performance of the MS analyzer.

Figure 1.3A shows profile spectra with different theoretical resolutions. If the resolution is too low, neighbouring peaks cannot be distinguished and isotopic resolution is lost. In this case, the average mass of a molecule will be determined. Nevertheless, even average spectra are suitable for determining the molecular weights of proteins from multiply charged ESI-MS signals⁴⁶  (http://www.bioprocess.org/esiprot/esiprot_form.php). To determine accurate masses, the peak shapes should be symmetrical. In the case of ToF analyzers, peak shapes and, therefore, the mass accuracies are affected by analyte concentration.⁴⁷ 

Comparing theoretical and experimental isotopic distributions is an efficient strategy for testing the validity of possible sum formulas from MS data, and, therefore, has been implemented in a variety of commercial and open source programs. With isotope distribution matching, likely chemical sum formulas can be derived even from spectra with modest mass accuracies and resolutions.^34,35,48 

The intensity of a mass signal can be used for quantification purposes; however, we should keep in mind that ions are observed rather than neutral molecules. Therefore, the abundant compounds in a sample may be invisible to MS if no ions are formed with the chosen ionization source. On the other hand, the levels of highly ionizable molecules can be overestimated.

Figure 1.3B displays an MS spectrum generated from experimental data. In short, Arabidopsis flowers were extracted with methanol and subjected to LC-MS. More information on this dataset and its evaluation is provided in Section 8.2. The data file contains 800 mass scans that were recorded during the chromatographic run. The presented scan (#245) corresponds to a retention time of 10.64 minutes. The strongest signal in the spectrum, with 309.0977 m/z, is referred to as the ‘base peak’.

This dataset is already centroided, but still contains many signals that later might be considered to be noise and removed. The density of peaks in this spectrum underlines the importance of high mass resolution for complex samples.

Fragmentation spectra are presented in the same format as full-scan MS spectra, but also report the fragmentation level and the precursor mass selected for fragmentation (e.g. MSMS 195, MS² 195, MS³ 195 → 82).

Chromatograms can be constructed from the ion intensities, and a total ion current chromatogram (TICC) plots the total ion current as a function of retention time. Figure 1.4A shows the base peak chromatogram (BPC) of the above-mentioned Arabidopsis flower extract analyzed by LC-MS. Only the most intense signal of each scan is considered for a BPC, and an extracted ion chromatogram (XIC) is finally constructed from the ion intensities in a chosen m/z window. An example is shown in Figure 1.4B, where only signals in the 736.2–736.4 m/z range were evaluated. LC-MS features are defined by their retention times and mass-to-charge ratios, both with their respective tolerances. The integrated ion intensities of such features are used for quantitative LC-MS analyses.

LC-MS analysis and data acquisition strategies are defined by the analytical scope of a project, and one needs to first choose between a targeted or an untargeted approach. In the first case, the compounds of interest are known and the complete workflow is directed towards the optimized detection of these analytes. However, the range of molecules of interest must also be restricted to something reasonable in the untargeted strategy (e.g., according to polarity, or m/z range, etc.).

On the basis of the properties of the expected molecules and the complexity of the sample, an extraction method (disintegration method, solvents, centrifugation, filtration, etc.) and separation method need to be defined, after which the ionization method, polarity, and a suitable mass analyzer need to be chosen appropriately.

An automated precursor-ion fragmentation process, also known as data-dependent acquisition (DDA), provides structural information for the identification of compounds. However, conventional DDA methods with MS and alternating MS² scans suffer from low efficiencies due to scan-cycle time limitations. Therefore, separate target-directed DDA experiments for the acquisition of fragmentation data provide superior results.⁴⁹  In addition, the additional data provided by MS full scan mode (i.e., without intermittent fragmentation scans) deliver a higher data density for quantitative analyses.

For the same reason, the sensitivities toward ions can be improved by selecting the ions of interest. In selected ion monitoring (SIM), only the defined ions of interest are recorded, thereby maximizing the sensitivities for those signals. In addition, only MSⁿ ions of a defined m/z precursor can be acquired in selected reaction monitoring (SRM). The strategy is referred to as consecutive reaction monitoring (CRM) when product ions from multiple sequential fragmentations are observed. Finally, SRM can be applied to multiple product ions from one or more precursor ions, resulting in multiple reaction monitoring (MRM).

Data are collected in either profile or centroid mode. As explained above, centroid spectra contain fewer data points, but are less bulky and data are acquired faster by the mass analyzer. The acquisition of centroid data is often preferable, especially in time-critical applications such as LC-MS (proteomics and metabolomics) and imaging. The collection of profile spectra is advantageous for producing spectra of the highest possible quality, as might be required for manual structural elucidation, among others.

Multiple sample types (organs, tissues, liquids, etc.), extraction methods (polar, non-polar, pH, etc.), and analytical strategies (MS, NMR, etc.) need to be combined to obtain a comprehensive picture of the physiology of a complex organism, such as human being or a plant.^50,51 

As discussed above, a well-defined analytical question greatly facilitates the experimental design while maintaining resource expenditure to within manageable limits. For example, the parallel profiling of phytohormones with an optimized sample preparation regime and the monitoring of suitable precursor‐to‐product ion transitions (LC-MS²) led to highly selective and sensitive results at reasonable costs, and revealed the dynamics of plant hormones in thermodormant seeds.⁵² 

Mass spectrometers are built by companies, so why shouldn't we analyze our data with the software they provide with their devices? Indeed, good commercial software exists for MS data analysis in (bio)chemistry and medicine. Integrated hardware–software solutions that are optimized for specific applications, easy to use, and validated, generate results of consistent quality. In addition, if questions arise, professional support is available. However, ‘all-in’ solutions are expensive, not only when first acquired, but also if additional software modules or software maintenance (updates/upgrades) are required. Frequently, only proprietary file formats are supported, which complicate data analyses with alternative programs (e.g., for data mining).⁵³  In the worst case, a commercial platform is not supported any more, which may lead to the ‘locked-in data syndrome’. The decision on the (dis)continuation of proprietary software is made by the owning companies based on the market situation. Therefore, using community data formats with long-term readability is highly recommended, even when using a commercial data processing platform.

In the worst-case scenario, a commercial platform becomes obsolete and is no longer supported, which may lead to ‘locked-in data syndrome’. The decision to (dis)continue proprietary software is made by the associated company based on the market. Therefore, the use of long-term-readable community data formats is highly recommended, even when a commercial data processing platform is used.

The development of open software is not driven by a financial motivation, but by academic interest. Often, scientists become developers because the available commercial software is not suitable for their objectives or is too costly. In the simplest case, a custom script is attached to commercial software to extend its functionality; this is not open software, but an ‘in-house solution’. Important requirements for releasing a program as open-source software (OSS) include the obligation to make the source program freely available to others and to allow others to modify and redistribute it. This is in stark contrast to proprietary programs that are usually delivered as compiled programs, and may not be copied, modified, or reverse engineered. The exact terms are defined in the end-user license agreements (EULAs). The Open Source Initiative defines ten OSS criteria and maintains a list of compatible licenses (https://opensource.org/).

OSS authors should release their code with a suitable license to define the terms and conditions of its use and copyright issues, among others. The community-driven development of OSS follows rules different to those used in industrial production. Yochai Benkler (2006) introduced the concept of peer production projects as large-scale, non-hierarchical, decentralized collaborations among multiple contributors. These are the result of the transition from an industrial economy to a networked information economy and made possible by 1. Diversity of Incentives, where creators of information goods are motivated by diverse reasons, and are not exclusively interested in financial rewards (copyright), and 2. Technological Shifts, in which the general availability of computers and internet access means that everyone is a potential author or distributor of information (democratization of publishing).⁵⁴ 

The advantages of open research have been impressively demonstrated by the production of an off-patent anti-parasitic drug, for which research accelerated after disclosure of the relevant information to the public, because more experts became involved in the project. Furthermore, the entire process is transparent and all project-related information is backed by a larger community and not maintained by only a few individuals.⁵⁵ 

The frequency of software updates and upgrades is another important distinctive feature. In OSS, bug fixes and function updates are released continuously and instantaneously following their implementation. On the other hand, new versions of commercial software are distributed less frequently. Since results can be affected by changes in the software, the corresponding software version (or snap-shot identifier) should always be documented.

From a socio-economical perspective, the use of OSS presents enormous potential. A current initiative of the Free Software Foundation (https://fsfe.org/) postulates the publication of code funded by public funds (https://publiccode.eu/). OSS may be installed on multiple computers without purchasing (additional) licenses, which leads to substantial cost reductions, especially in academic institution with multiple users. In addition, the program code can be inspected. As an example, Dr Chris Rath used my ESIprot program (https://bitbucket.org/lababi/esiprot) in his class and encountered a bug that I was able to quickly fix. Consequently, all participants profit from the OSS concept.

Table 1.3 compares the typical characteristics of commercial and open source software.

The priorities in an industrial or service laboratory are different to those of academia and education. Software stability, result reproducibility, and regulatory compliance are mandatory. Program malfunction or the loss of data following updates or third-party changes (e.g., public web API) are unacceptable; consequently, companies such as LabKey (https://www.labkey.com/) offer different versions of their software platforms. The free community server edition offers core open source features and is supported by a user forum, whereas professional/enterprise versions include proprietary modules and additional support. Such ‘multi-licensing’ models provide an excellent compromise between the dynamics of peer software production and those of a productive environment and the requirements of a productive environment.