Computational Expansion of High-Resolution-MSn Spectral Libraries Authors: Brandon Y. Lieng, Andrew T. Quaile, Xavier Domingo-Almenara, Hannes L. Röst, J. Rafael Montenegro-Burke Published: 2023 ABSTRACT Commonly, in MS-based untargeted metabolomics, some metabolites cannot be confidently identified due to ambiguities in resolving isobars and structurally similar species. To address this, analytical techniques beyond traditional MS2 analysis, such as MSn fragmentation, can be applied to probe metabolites for additional structural information. In MSn fragmentation, recursive cycles of activation are applied to fragment ions originating from the same precursor ion detected on an MS1 spectrum. This resonant-type collision-activated dissociation (CAD) can yield information that cannot be ascertained from MS2 spectra alone, which helps improve the performance of metabolite identification workflows. However, most approaches for metabolite identification require mass-to-charge (m/z) values measured with high resolution, as this enables the determination of accurate mass values. Unfortunately, high-resolution-MSn spectra are relatively rare in spectral libraries. Here, we describe a computational approach to generate a database of high-resolution-MSn spectra by converting existing low-resolution-MSn spectra using complementary high-resolution-MS2 spectra generated by beam-type CAD. Using this method, we have generated a database, derived from the NIST20 MS/MS database, of MSn spectral trees representing 9637 compounds and 19386 precursor ions where at least 90% of signal intensity was converted from low-to-high resolution. The metabolome is associated with a multitude of human diseases, such as diabetes mellitus, inflammation, and cancer.1 Hence, untargeted metabolomics aims to detect and identify as many metabolites comprising a metabolome as possible. To this end, mass spectrometry (MS) is often applied. However, in MS-based untargeted metabolomics experiments, m/z values of ions detected at MS1 can often correspond to multiple possible metabolites.2 To resolve such ambiguities, collision-activated dissociation (CAD) is used to fragment ions (herein “precursor ions”) and generate MS2 fragmentation spectra, which represent “fingerprints” of precursor ions.3−7 Yet, even with their MS2 spectra, most metabolites detected in metabolomics experiments still cannot be confidently annotated.8,9 Ambiguities in identification often arise due to isobaric or structurally similar species that produce similar fragmentation spectra. To address this, other analytical techniques are often employed alongside MS2 analysis to generate additional structural information. For example, separation techniques such as liquid chromatography and ion mobility spectrometry coupled to MS can determine a metabolite’s retention time and collisional cross-section, respectively.9 Similarly, different fragmentation regimes can be applied to generate complementarily informative fragment ions from the same precursor ion.10 For example, under a resonant-type CAD regime, only ions of a specific m/z are activated. In contrast, under a beam-type CAD regime, all ions, including fragment ions, are activated. Consequently, different sets of fragment ions can be observed between fragmentation spectra of the same precursor ion generated by either resonant- or beam-type CAD. Despite the availability of these complementary approaches, metabolite identification remains a difficult task. MSn fragmentation applies recursive rounds of fragmentation to a precursor ion and its fragment ions. In an MSn experiment, an MSn spectrum is generated by selecting ions detected on an MSn–1 spectrum for isolation and fragmentation. This produces a “spectral tree” describing the hierarchical relationships between MSn spectra succeeding a particular MS1 spectrum.4,11,12 An example MSn experiment and its corresponding spectral tree are shown in Figure 1. Since MSn experiments reveal hierarchical relationships between fragment ions, MSn spectra encode information about metabolite structures more accurately than MS2 spectra on their own, suggesting that MSn fragmentation may help improve the performance of current metabolite identification workflows.13,14 Illustration of an MSn fragmentation experiment in a linear ion trap (left) and the corresponding MSn spectral tree (right). Note that the term MSn typically refers to cases where n ≥ 3. Metabolite identification workflows often involve the task of mass decomposition, where molecular formulae are assigned to fragment mass values, requiring mass values measured with high accuracy to reduce ambiguity.2,15,16 To achieve high mass accuracy, spectra must be collected at high resolution. Unfortunately, MSn spectra acquired at high resolution are relatively rare in current publicly available spectral databases. The lack of high-resolution-MSn fragmentation spectra available in public databases can be partly attributed to the fact that MSn fragmentation requires resonant-type CAD available only via low-resolution and, consequently, low-accuracy (±0.15 m/z error, equivalent to 375 ppm error for an ion with m/z = 400) ion trap instrumentation.17 On the other hand, high-resolution mass analyzers such as Orbitraps and time-of-flight-type instruments are designed with beam-type CAD capabilities.18,19 Thus, most MSn experiments have historically been performed using instruments that are incapable of high-resolution measurements. While Fourier transform-ion cyclotron resonance (FT-ICR) instruments can acquire high-resolution-MSn data, they are rarely used for untargeted metabolomic profiling due to their high cost and slow acquisition speeds, which are incompatible with ultra-high-performance LC time scales. Recent developments in MS instrumentation have yielded hybrid mass spectrometers that combine multiple mass analyzers and fragmentation modules into a single instrument. Hybrid MS instruments enable the acquisition of high-resolution-MSn spectra, since hybrid instruments containing both a linear ion trap (LIT) and an Orbitrap can perform MSn fragmentation in the LIT before transferring the resulting fragment ions to the Orbitrap for high-resolution mass analysis. However, generating a spectral library requires the collection of thousands of reference spectra—a laborious and resource-intensive process—which explains the limited availability and accessibility of spectra in current spectra repositories such as GNPS, METLIN, MoNA, and mzCloud (www.mzcloud.org).20−23 To leverage the additional information MSn experiments can yield using current metabolite identification tools, a high-resolution-MSn spectral database must be generated. Here, we describe a computational method to convert low-resolution peaks found throughout existing low-resolution-MSn spectra in the NIST20 MS/MS database to a higher resolution using complementary high-resolution-MS2 data generated by beam-type CAD (Figure 2). To achieve this, we first identified precursor ions represented by various types of spectra in NIST20, which we used to demonstrate the validity of two underlying assumptions of the conversion method for a given precursor ion: (1) MSn fragmentation by resonant-type CAD produced fragment ions that are also observed under MS2 fragmentation by beam-type CAD and (2) fragmentation spectra acquired using a linear ion trap were comparable to those acquired using an Orbitrap. After verifying these assumptions, we validated our conversion workflow using a subset of precursor ions in NIST20 for which both low- and high-resolution-MSn spectra were available. Finally, we applied the peak resolution conversion method to all precursor ions in NIST20 represented by both low-resolution-MSn and high-resolution-MS2 spectra generated by beam-type CAD, resulting in a ∼9-fold expansion of high-resolution-MSn spectra. (a) Different combinations of fragmentation methods and mass analyzers were used to acquire three types of spectra in NIST20. (b) Unique precursor ions represented in NIST20 by each of the three types of spectra were identified. 27159 precursor ions were represented by both LR-RES-MSn and HR-BEAM-MS2 spectra (dotted outline). (c) To leverage these overlapping data, we proposed a peak resolution conversion method whereby information from an LR-RES-MSn spectrum is combined with information from HR-BEAM-MS2 spectra of the same precursor ion to produce a converted-MSn spectrum. (d) In this method, LR-RES-MSn peaks are converted to high-resolution peaks using high-resolution m/z values found on HR-BEAM-MS2 spectra. (e) In this study, we first checked assumptions underlying the method, validated the method using a subset of 1404 precursor ions for which HR-RES-MSn spectra were already available in NIST20, and then applied the method to the subset of 27159 precursor ions eligible for conversion. Experimental Section NIST20 MS/MS Database Precursor ions represented in NIST20 by high-resolution-MS2 spectra generated by beam-type CAD (HR-BEAM-MS2 spectra), high-resolution-MSn spectra generated by resonant-type CAD (HR-RES-MSn spectra), and low-resolution-MSn spectra generated by resonant-type CAD (LR-RES-MSn spectra) were identified (Figure 2a,b).12 Here, the term “precursor ion” refers to a unique pairing of a compound and adduct/neutral loss. HR- and LR-RES-MSn spectra in NIST20 were acquired at a normalized collision energy (NCE) of 35% and at MS levels MS2 to MS4. HR-BEAM-MS2 spectra were acquired at a resolution of R = 30000 (full width at half-maximum @ 200 m/z) and across various NCEs.12 NIST20 contains HR-RES-MSn, LR-RES-MSn, and HR-BEAM-MS2 spectra for 1081, 11797, and 26262 compounds, respectively. Matching Peaks Peaks were matched between spectra using a Python script and error tolerances of ±0.001 m/z when comparing HR-RES-MSn to HR-BEAM-MS2 spectra and ±0.1 m/z when comparing HR-RES-MSn to LR-RES-MSn spectra. Peaks on HR-RES-MSn were rounded down to one decimal place prior to comparisons with LR-RES-MSn spectra. Percentages of peaks matched within HR- or LR-RES-MSn spectral trees were calculated by determining the proportion of peaks within each tree that were matched to peaks on corresponding spectra. To calculate percentages of signal intensity matched, matched peaks were weighted by their relative intensities prior to calculating proportions of peaks matched. Spectral Similarity Scores Cosine similarity scores were computed using the matchms Python package.24 An error tolerance of ±0.001 m/z was used when computing scores between HR-RES-MSn spectra from NIST20 and converted-MSn spectra, whereas an error tolerance of ±0.005 m/z was used when computing scores between HR-RES-MSn spectra experimentally acquired in-house and converted-MSn spectra. Peak Resolution Conversion A computational method was implemented to convert the resolution of peaks found throughout LR-RES-MSn spectra from low-to-high using the m/z values of matched peaks from corresponding HR-BEAM-MS2 spectra (Figure 2c). Note that the conversion of a peak from low-to-high resolution also implicitly converts the mass accuracy of the peak from low-to-high. For precursor ions where both LR-RES-MSn and HR-BEAM-MS2 spectra were available, LR peaks throughout the LR-RES-MSn spectral tree were matched to HR peaks on corresponding HR-BEAM-MS2 spectra. If a matching peak was found, the m/z value of the LR peak was replaced with the m/z value of the HR peak, effectively converting the LR peak to a higher resolution (Figure 2d). When multiple HR peaks were found within the ±0.1 m/z window for an LR peak, the LR m/z value was heuristically converted using the m/z value associated with the most intense HR peak. HR-RES-MSn Spectra Generation HR-RES-MSn spectra were acquired using standard reference materials. Dry powders of quinine hydrochloride, verapamil hydrochloride, glafenine hydrochloride, ouabain octahydrate, and colchicine (Sigma-Aldrich, St. Louis, MO) were individually solubilized in LC-MS-grade methanol and water (1:1, v/v; Fisher Chemical, Hampton, NH) to a concentration of 1 μg/mL. Each reference solution was then directly infused at 5 μL/min into an Orbitrap IQ-X Tribrid mass spectrometer (Thermo Scientific, Waltham, MA). HR-RES-MSn fragmentation was performed in the LIT with parameters matching those used in NIST20 (NCE of 35% and resolution of R = 30000).12 Spray parameters were set to their default values in Tune based on the 5 μL/min infusion rate. Spectra used in spectral comparisons were each comprised of ten individual scans, averaged using the SpectraMerger tool available in OpenMS.25 After merging, noise peaks (<10% relative intensity) were removed from MS2 and MS3 spectra. Similarly, peaks with less than 20% relative intensity were removed from MS4 spectra. Generation of Fragmentation Trees A fragmentation tree for verapamil [M + H]+ was obtained using its LR-RES-MSn spectral tree from NIST20 and the corresponding converted-MSn spectral tree. In this fragmentation tree, a node was created for the base peak of each spectrum in the spectral tree. Each spectrum in the spectral tree corresponded to exactly one node in the fragmentation tree; thus, MSn → MSn+1 relationships in the spectral tree were directly translated as edges between nodes. Since all peaks included as nodes in the fragmentation tree were converted from low-to-high resolution using the conversion method, each node was associated with a pair of low-accuracy and converted (high-accuracy) m/z values. Generation of Molecular Formulas The ChemCalc Web Service was used to perform all mass decompositions described.26 Low-accuracy and converted m/z values were decomposed using error tolerances of ±0.1 and ±0.001 m/z units, respectively. To generate molecular formulae for the verapamil [M + H]+ fragmentation tree example, the elements C[0–27], H[0–39], N[0–2], and O[0–4] were allowed based on the molecular formula of the precursor ion. Possible molecular formulae for each node were constrained to subformulae of the respective parent node. To generate molecular formulae for the 1000 MS3 peaks sampled to assess the effect of conversion on mass decomposition, the elements C[0–105], H[0–180], N[0–20], O[0–75], P[0–5], S[0–5], Br[0–5], F[0–15], Cl[0–10], and B[0–1] were allowed. Results and Discussion NIST20 Spectral Library First, precursor ions represented by HR-RES-MSn, LR-RES-MSn, and HR-BEAM-MS2 spectra in NIST20 were identified. Notably, some precursor ions were represented by multiple types of spectra and, most interestingly, 27159 precursor ions were represented by both LR-RES-MSn and HR-BEAM-MS2 spectra (Figure 2b). These precursor ions represented cases where our peak resolution conversion method could be applied. Instruments used to acquire spectra in these subsets of NIST20 are listed in Supporting Information Table S1. Checking Assumptions of the Conversion Method Before the peak resolution conversion method was implemented, two assumptions underlying the method had to be tested. First, for a given precursor ion, fragment ion peaks found on RES-MSn spectra are also found on BEAM-MS2 spectra. This was important to test as resonant-type and beam-type CAD differ with respect to the frequency of collisions that precursor ions are subjected to under either regime. This could lead to different distributions of internal energy within precursor ions and the formation of different fragment ions.10 Second, fragment ion peaks found on LR-MSn spectra are also found on HR-MSn spectra—it was important to ensure that ions detected in the LIT are also transferred to and detected in the Orbitrap. To test the first assumption, HR-RES-MSn and HR-BEAM-MS2 spectra from NIST20 representing the same precursor ion were compared (Figure 3a). Peaks from HR-RES-MSn spectra were matched to peaks on corresponding HR-BEAM-MS2 spectra. Following this, the percentage of peaks and intensity matched throughout HR-RES-MSn spectra was calculated. (a) Peaks in HR-RES-MS2 spectra were matched to corresponding HR-BEAM-MSn spectra. Peaks shown in black are matched and peaks shown in red are unmatched. (b) The distribution of percentages of signal intensity matched within each precursor ion’s HR-RES-MSn spectral tree to a set of corresponding HR-BEAM-MS2 spectra. The red dashed line denotes the median value. (c) Peaks in LR-RES-MSn spectra were matched to HR-RES-MSn spectra that had an identical MSn fragmentation pathway. (d) The distribution of percentages of signal intensity matched within each precursor ion’s LR-RES-MSn spectral tree. Across all 1957 precursor ions for which both HR-RES-MSn and HR-BEAM-MS2 spectra were available (yellow and gray sections in Figure 2b), 77.41% of HR-RES-MSn peaks (137372 matched/166115 total) were matched to peaks found throughout corresponding HR-BEAM-MS2 spectra. Although many peaks were not matched from HR-RES-MSn to HR-BEAM-MS2 spectra, weighing peaks by their intensity revealed that the majority (94.48%) of total HR-RES-MSn signal intensity was matched to an HR-BEAM-MS2 peak. Next, the distribution of the percentages of signal intensity matched within each precursor ion’s HR-RES-MSn spectral tree was visualized (Figure 3b). The median percentage of signal intensity throughout a precursor ion’s HR-RES-MSn spectral tree that was matched to a peak in the same precursor ion’s HR-BEAM-MS2 spectra was 99.33%. Taken together, these data validate the first assumption, showing that HR-RES-MSn fragments are consistently present on HR-BEAM-MS2 spectra. To test the second assumption, LR-RES-MSn and HR-RES-MSn spectra from NIST20 representing the same precursor ion and MSn fragmentation pathway were compared (Figure 3c). Across all 1469 precursor ions for which both LR- and HR-RES-MSn spectra were available (cyan and gray sections in Figure 2b), 4192 pairings of LA and HA spectra that represented the same MSn fragmentation pathways were formed. Peaks from LR-RES-MSn spectra were then matched to peaks on paired HR-RES-MSn spectra. The percentage of peaks and signal intensity matched throughout LR-RES-MSn spectra was calculated. Across all comparisons, 54.37% of LR-RES-MSn peaks (68688 matched peaks/126332 total peaks) were matched to corresponding peaks found throughout paired HR-RES-MSn spectra. When peaks were weighted by their signal intensity, 91.81% of LR-RES-MSn signal intensity was matched to a corresponding HR-RES-MSn peak. In keeping with the previous analysis, the distribution of the percentages of signal intensity matched within each precursor ion’s LR-RES-MSn spectral tree was visualized (Figure 3d). The median percentage of signal intensity throughout a precursor ion’s LR-RES-MSn spectral tree that was matched to a peak found in paired HR-RES-MSn spectra was 97.70%. These results validate the second assumption, as most of the signal intensity present throughout a precursor ion’s LR-RES-MSn spectra was usually found throughout its HR-RES-MSn spectra. Validation of the Conversion Method Since the above analyses demonstrated the validity of both assumptions underlying the peak resolution conversion method, the method was implemented and validated using a subset of 1404 precursor ions in NIST20 for which LR-RES-MSn, HR-BEAM-MS2, and HR-RES-MSn spectra were available (gray section in Figure 2b). Peaks in LR-RES-MSn spectra were converted from low-to-high resolution using the implemented method. The percentage of signal intensity converted within each LR-RES-MSn spectral tree was then calculated. The median percentage of signal intensity within each precursor ion’s LR-RES-MSn spectral tree that was matched to an HR-BEAM-MS2 peak and converted to high resolution was 97.34% (Figure 4a). Examples of available HR-BEAM-MS2 spectra and converted-MSn spectra for the precursor ion avocadene acetate [M + H–H2O]+ are shown in Figure 4d,e, respectively. (a) The distribution of percentages of signal intensity converted from LA to HA within each LR-RES-MSn spectral tree used to evaluate the method. (b) The distribution of cosine similarity scores calculated between pairs of converted-MSn and HR-RES-MSn spectra that represent the same MSn fragmentation pathway. The red dashed line denotes the median value. (c) The structure of avocadene acetate. (d) Composite spectrum of HR-BEAM-MS2 spectra available for avocadene acetate [M + H–H2O]+. (e) Converted-MSn spectra for avocadene acetate [M + H–H2O]+. Peaks converted from low-to-high resolution are shown in black, whereas unconverted peaks are shown in red. Annotations denote each spectrum’s MSn fragmentation pathway and percentage of signal intensity converted. (f) An example showing the conversion of the LR peak denoted by the red star in (e) to HR. Here, HR-BEAM-MS2 spectra were searched for a peak with an m/z value in the range (149.0, 149.2). Three candidate m/z values were found, and the one associated with the peak with the highest relative intensity (m/z = 149.1325) was used to convert the LR peak to HR. Additionally, since HR-RES-MSn spectra were available in NIST20 for this subset of precursor ions, they were compared against converted-MSn spectra—between 4012 pairs of converted- and HR-RES-MSn spectra that represented the same precursor ion and MSn fragmentation pathway, cosine spectral similarity scores were computed. The median spectral similarity score between converted-MSn spectra and their corresponding HR-RES-MSn spectra was 0.99 (Figure 4b). In most cases, multiple HR peaks were found within the ±0.1 m/z window for each converted LR peak and the m/z value of the most intense HR peak was used for conversion (Supporting Information Figures 1 and 2). However, the fact that median spectral similarity scores were high suggested that, in most cases, the most intense peaks were usually converted to the correct HR m/z value or to one within ±0.001 m/z of the correct value (Figure 4b). These results demonstrate that the peak resolution conversion method performed well, as most of the signal intensity throughout LR-RES-MSn spectral trees in this subset was converted to HR. Furthermore, these converted-MSn spectra shared high cosine similarity scores when compared to HR-RES-MSn spectra using a peak matching tolerance suitable for the comparison of high-resolution spectra. Generation of a Converted-MSn Spectral Database Following validation, the peak resolution conversion method was next applied to all 27203 LR-RES-MSn spectral trees in NIST20 for which HR-BEAM-MS2 spectra of the same precursor ions were also available (pink and gray sections in Figure 2b). This generated a database of converted-MSn spectral trees representing 11562 compounds and 27159 precursor ions (44 precursor ions are represented by two spectral trees each). The proportion of signal intensity converted within each LR-RES-MSn spectral tree was then calculated. The median percentage of signal intensity within each precursor ion’s LR-RES-MSn spectral tree that was matched to an HR-BEAM-MS2 peak and converted to high resolution was 97.20% (Figure 5a). This is comparable to the median percentage of signal intensity matched and converted within each spectral tree during validation of the method (97.34%; Figure 4a). Spectral trees where more than 90% of LR-RES-MSn intensity was converted from low-to-high resolution were considered as successfully converted to high resolution. Using this cutoff, 9637 compounds and 19386 precursor ions were represented by converted-MSn spectral trees in the converted database. This corresponds to an 8.9-fold increase over the 1081 compounds and a 9.4-fold increase over the 2057 precursor ions that were previously represented by HR-RES-MSn spectral trees in NIST20 (Figure 5b). (a) Distribution of percentages of signal intensity converted from LR to HR within each LR-RES-MSn spectral tree. (b) Counts of compounds and precursor ions represented by high-resolution-MSn spectra in NIST20 before (light green area) and after (light green area plus dark green area) converting the resolution of LR-RES-MSn peaks where possible. Areas are proportional to counts. Comparison of Experimentally Acquired Spectra to Converted-MSn Spectra To demonstrate that experimentally acquired, high-resolution-MSn spectra would produce high spectral similarity scores to spectra in the converted-MSn database, spectral similarity was assessed between high-resolution-MSn spectra acquired in-house and a subset of converted-MSn spectra. Here, HR-RES-MSn spectra were collected for five compounds represented by converted-MSn spectral trees (i.e., >90% intensity converted) and that were not previously represented by HR-RES-MSn spectra in NIST20 (pink section in Figure 2b). In 22 of 31 comparisons, MSn spectra acquired at high resolution shared high spectral similarity scores (>0.9) with converted-MSn spectra from NIST20, even when using a stringent error tolerance (±0.005 m/z) to match peaks. These observations are consistent with spectral similarity scores computed between HR-RES-MSn already in NIST20 and converted-MSn spectra (Figure 4b). An example spectral tree of mirror plots and their associated similarity scores is shown in Figure 6a for one of the analyzed precursor ions, verapamil [M + H]+ (Figure 6b). Mirror plots and similarity scores for the other four compounds are shown in Supporting Information Figure 3. This is a notable observation, as spectral similarity scores between reference spectra and experimentally acquired spectra are often considered when annotating detected metabolites with putative identities. (a) Spectral tree of mirror plots comparing high-resolution-MSn spectra acquired in-house (above the x-axis) to converted-MSn spectra from NIST20 (below the x-axis). Peaks shown in red were not converted and remain as low-resolution peaks. Annotations denote MSn fragmentation pathways and cosine spectral similarity scores calculated using the matchms package and a peak matching tolerance of ±0.005 m/z units.24 (b) The precursor ion being fragmented in this example is verapamil [M + H]+. (c) Number of candidate molecular formulae for each node in a fragmentation tree of the precursor ion verapamil [M + H]+. Each node corresponds to a spectrum in the original spectral tree (shown in a) and its base peak. Values along edges represent precursor m/z values. (d) Percent reductions in the numbers of candidate molecular formulae explaining 1000 sampled MS3 peaks, depending on whether the LA or converted m/z value was decomposed. Conversion of a peak from LA to HA reduced the number of candidates by roughly two orders of magnitude in most cases. Generation of Molecular Formulas for Low-Accuracy and Converted m/z Values To demonstrate that the conversion of peak resolution and mass accuracy facilitates mass decomposition, mass decomposition was assessed using two approaches—one using an example precursor ion from NIST20, and the other using a random sample of converted peaks. In the first approach, a fragmentation tree for the precursor ion verapamil [M + H]+ was manually generated and mass decompositions were performed to annotate each tree node with putative molecular formulae explaining the node’s pair of low-accuracy and converted m/z values. Fragmentation trees are useful for metabolite identification, as they model the fragmentation reactions that produced a fragmentation spectrum.27,28 Using low-accuracy m/z values and an error tolerance of ±0.1 Da, all nodes in the manually generated verapamil [M + H]+ fragmentation tree were decomposed to multiple candidate molecular formulae, including the MS2 precursor m/z, which decomposed to 31 possible molecular formulae (Figure 6c). However, using converted m/z values and an error tolerance of ±0.001 Da, all nodes were explained by exactly one molecular formula. In this example, conversion of an LR-RES-MSn spectral tree from NIST20 enabled the unambiguous decomposition of peak m/z values to molecular formulae. Mass decomposition was also assessed in a more general fashion by sampling 1000 converted MS3 peaks and computing possible molecular formulae for each peak’s low-accuracy and converted m/z values. As expected, the number of candidate decompositions of converted m/z values was always smaller than the number of candidate decompositions of low-accuracy m/z values. On average, conversion of a low-accuracy m/z value allowed for the reduction of the size of the candidate decomposition list by 98.88% or by roughly two orders of magnitude, demonstrating again that converted-MSn spectra facilitate the task of mass decomposition (Figure 6d). Conclusions Here, we present a computational framework to convert low-resolution peaks found throughout MSn spectra in NIST20 to a higher resolution using complementary MS2 spectra generated by beam-type CAD fragmentation. We first demonstrated that fragmentation of precursor ions under a resonant-type CAD regime tends to generate fragment ions that are also observed under a beam-type CAD regime. We also showed that most signals observed throughout low-resolution-MSn spectra are also found in corresponding high-resolution-MSn spectra of the same precursor ion and MSn fragmentation pathway. Spectral similarity scores calculated between MSn spectra collected with high resolution and corresponding converted-MSn spectra revealed that, in most cases, such spectra were highly similar. Importantly, our computational methodology led to the expansion of the coverage of compounds and precursor ions represented by high-resolution-MSn spectra in the NIST20 database by ∼9-fold, using either measure, and did not require the generation of any additional spectral data using reference standard materials, thus saving a considerable amount of material costs and research personnel hours. Our method is available as a Python script, accessible at www.github.com/montenegroburkelab/rescon, allowing users with access to the NIST20 spectral library to generate the converted-MSn library. The script accepts input fragmentation spectra, subsets for the relevant spectra, converts the spectra, and outputs converted-MSn spectra in the MSP format. Furthermore, this method is generalizable to other LR-RES-MSn spectra not in the NIST20 database.