Testverfahren für automatische Spektreninterpretationsmethoden

Automatic spectra interpretation methods (e. g. pattern recognition methods for the interpretation of mass spectra) should be characterized by suitable criteria of quality. These criteria may be obtained by testing interpretation methods with a random sample of spectra, but should be independent from probabilities of classes in this sample, or should refer to a sample with equal probabilities of classes. In this paper, mathematical formulae for such objective criteria of quality are given. For example, individual predictive abilities for the classes and maximum information are appropriate to characterize and compare interpretation methods from different authors.

---

---

Mit 2 Abbildungen     

Establishment of correlations between the structure and reactivity of molecules in the gas phase based on information theory

A general approach to revealing correlations between the structure of molecules and their reactivity in fragmentation processes under electron impact conditions based on the use of generalized structural and mass spectral characteristics is suggested. The characteristics were obtained using information theory, molecular graphs, and absolute reaction rates. Information topological indices of molecular graphs were used as generalized structural characteristics of molecules. They are a quantitative measure of the structural complexity of molecules and are expressed in information units. The gas-phase process of fragmentation of molecules under electron impact was used as a general reaction series for all volatiles. In terms of information theory, the mass spectrum represents the distribution of probabilities of the formation of ions of each type, and the information entropy of this distribution appears to be an integral characteristic of the reactivity of a molecule during fragmentation under electron impact in the gas phase. Using organic and organometallic compounds of several classes (ferrocene derivatives, arylsilanes, aromatic azo compounds,etc.) as examples, linear correlations between the information indices of the mass spectra and the information topological indices of the appropriate molecular graphs or electronic parameters of molecules have been found, which testifies that the approach suggested is adequate.Translated fromIzvestiya Akodemii Nouk. Seriya Khimicheskaya, No. 11, pp. 2683–2688, November, 1996.     

Predicting cis/trans structure of alkene based on infrared spectra.

The application of chemometrics to analyze the information of the cis/trans structure of alkenes in infrared spectra (IR) is introduced. For data from the OMNIC IR spectral database, two feature selection methods, Fisher ratios and genetic algorithm-partial least squares (GA-PLS), and two classification methods, support vector machine (SVM) and probabilistic neural network (PNN), have been used to obtain optimization classifiers. At last, some spectra from other IR databases are used to evaluate the optimization classifiers. It has been demonstrated that both the SVM and PNN optimization classifiers could give preferable predictive results about the cis and trans structures of alkene.     

Predictive activity profiling of drugs by topological-fragment-spectra-based support vector machines

Aiming at the prediction of pleiotropic effects of drugs, we have investigated the multilabel classification of drugs that have one or more of 100 different kinds of activity labels. Structural feature representation of each drug molecule was based on the topological fragment spectra method, which was proposed in our previous work. Support vector machine (SVM) was used for the classification and the prediction of their activity classes. Multilabel classification was carried out by a set of the SVM classifiers. The collective SVM classifiers were trained with a training set of 59,180 compounds and validated by another set (validation set) of 29,590 compounds. For a test set that consists of 9,864 compounds, the classifiers correctly classified 80.8% of the drugs into their own active classes. The SVM classifiers also successfully performed predictions of the activity spectra for multilabel compounds.     

The use of MS classifiers and structure generation to assist in the identification of unknowns in effect-directed analysis

Structure generation and mass spectral classifiers have been incorporated into a new method to gain further information from low-resolution GC-MS spectra and subsequently assist in the identification of toxic compounds isolated using effect-directed fractionation. The method has been developed for the case where little analytical information other than the mass spectrum is available, common, for example, in effect-directed analysis (EDA), where further interpretation of the mass spectra is necessary to gain additional information about unknown peaks in the chromatogram. Structure generation from a molecular formula alone rapidly leads to enormous numbers of structures; hence reduction of these numbers is necessary to focus identification or confirmation efforts. The mass spectral classifiers and structure generation procedure in the program MOLGEN-MS was enhanced by including additional classifier information available from the NIST05 database and incorporation of post-generation ‘filtering criteria’. The presented method can reduce the number of possible structures matching a spectrum by several orders of magnitude, creating much more manageable data sets and increasing the chance of identification. Examples are presented to show how the method can be used to provide ‘lines of evidence’ for the identity of an unknown compound. This method is an alternative to library search of mass spectra and is especially valuable for unknowns where no clear library match is available.     

Advantages of high-resolution and high-mass range magnetic-sector mass spectrometry for electrospray ionization

Electrospray ionization (ESI) mass spectra have been measured on a magnetic-sector double-focusing mass spectrometer for a number of proteins and peptides. It is pointed out how in theory raising the mass resolution of a mass spectrometer from 800–1000 to 2400–3000 significantly increases the precision with which the envelope of isotopic peaks of a protein ion (or other organic ion) can be defined, particularly at higher masses. Better definition of the isotopic envelope ought to lead to higher precision in the experimental determination of molecular mass, which has been demonstrated. It is shown how ESI mass spectra of high-mass molecules are significantly less congested at higher m/z values, so that for these molecules (RMM > 40 000) there is an advantage in being able to record peaks at higher m/z values (m/z > 2000) representing ions with fewer charges. Fragmentation of a small peptide in the ESI source has been found to provide sequence information.     

Classification of time-of-flight secondary ion mass spectrometry spectra from complex Cu–Fe sulphides by principal component analysis and artificial neural networks

Artificial neural network (ANN) and a hybrid principal component analysis-artificial neural network (PCA-ANN) classifiers have been successfully implemented for classification of static time-of-flight secondary ion mass spectrometry (ToF-SIMS) mass spectra collected from complex Cu–Fe sulphides (chalcopyrite, bornite, chalcocite and pyrite) at different flotation conditions. ANNs are very good pattern classifiers because of: their ability to learn and generalise patterns that are not linearly separable; their fault and noise tolerance capability; and high parallelism. In the first approach, fragments from the whole ToF-SIMS spectrum were used as input to the ANN, the model yielded high overall correct classification rates of 100% for feed samples, 88% for conditioned feed samples and 91% for Eh modified samples. In the second approach, the hybrid pattern classifier PCA-ANN was integrated. PCA is a very effective multivariate data analysis tool applied to enhance species features and reduce data dimensionality. Principal component (PC) scores which accounted for 95% of the raw spectral data variance, were used as input to the ANN, the model yielded high overall correct classification rates of 88% for conditioned feed samples and 95% for Eh modified samples.     

Mixed ligand complexes of β-diketonates: synthesis,characterization, and FAB mass spectral analysis

Complexes of the type MLL′ · nB (where M = Ni(II) and Cu(II); LH and L′H = 2,4-pentanedione (acacH), 1-phenyl-1,3-butanedione (bacH), and 1,3-diphenyl-1,3-propanedione (dbmH); n = 0 to 2 and B = water or pyridine) have been synthesized and characterized. IR spectra are consistent with uninegative bidentate ligands. Magnetic moments and electronic spectral studies reveal high-spin octahedral geometry for nickel(II) complexes and distorted octahedral stereochemistry for copper(II) complexes. Frozen chloroform solution ESR spectra of the copper(II) complexes display significant Jahn–Teller distortion and dimeric behavior of the complexes in solution. FAB mass spectra of the copper(II) complexes also exhibit peaks corresponding to dimers. Molecular, pseudo-molecular, dimeric pseudo-molecular, and fragment ion peaks in unit resolution mass spectra have been identified with the help of their isotope distribution pattern expected due to natural abundances of the ⁶³Cu and ⁶⁵Cu isotopes. All the FAB mass spectral peaks from the fragment ions containing copper have been interpreted on the basis of isotope distribution pattern.     

Beitrag zur Interpretation von Massenspektren mit Hilfe adaptiver, linearer Klassifikatoren

Zusammenfassung Substanzklassen und chemische Strukturen niedrigmolekularer, organischer Verbindungen können mit Hilfe linearer, binärer, adaptiver Klassifikatoren aus den niedrig aufgelösten Massenspektren mit einer gewissen Wahrscheinlichkeit ermittelt werden.Mit einem Trainingsset von 250 Massenspektren wurden für 31 Substanzklassen und chemische Strukturelemente derartige Klassifikatoren berechnet. Weitere 250 Massenspektren dienten zur Prüfung der KlassifizierungsfähigkeitP für unbekannte Spektren.Als günstigste Vorbehandlung der Spektrendaten wurde die logarithmische Intensitätstransformation und die Normierung auf den Basispeak gefunden.Wenn eine Substanzklasse im Trainingsset nur wenige Vertreter hat, dann istP für Spektren dieser Klasse meist sehr klein und der Klassifikator unbrauchbar.Bei kritischer Untersuchung erwiesen sich nur insgesamt 10 Substanzklassen und chemische Strukturelemente für diese Methode der automatischen Massenspektreninterpretation als geeignet.

---

Contribution to the interpretation of mass spectra with adaptive linear classifiers

Molecular structures of organic substances with low molecular weight can be determined with some probability from low resolution mass spectra by linear, binary, adaptive classifiers. 250 mass spectra were used as a training set to compute such classifiers for 31 different molecular structures; other 250 mass spectra were used to test the predictive abilityP. The best preprocessing of the spectral data was the logarithmic transformation of the intensities and normalization to the base peak. If the training set contains only some members of a certain class,P is very low for spectra of this class and the trained classifier is not useful. After critical examination only 10 molecular structures remained, which are suitable for this method of automatic mass spectra interpretation.

     

Neue Entscheidungsvektoren für Massenspektren zur Klassifizierung von 10 chemischen Strukturen

Adaptive, linear classifiers (decision vectors), suitable for automatic mass spectra interpretation, were calculated from 500 low resolution mass spectra. The convergence rate of the vector training could be enlarged by using previously trained vectors as initial decision vectors. Some relationships between decision vectors and mass spectrometric fragmentation rules are discussed.     

A comparison of electron ionization mass spectra obtained at 70 eV,low electron energies,and with cold EI and their NIST library identification probabilities

Electron ionization (EI) mass spectra of 46 compounds from several different compound classes were measured. Their molecular ion abundances were compared as obtained with 70‐eV EI, with low eV EI (such as 14 eV), and with EI mass spectra of vibrationally cold molecules in supersonic molecular beams (Cold EI). We further compared these mass spectra in their National Institute of Standards and Technology (NIST) library identification probabilities. We found that

1. Low eV EI is not a soft ionization method, and it has little or no influence on the molecular ion relative abundances for large molecules and those with weak or no molecular ions.
2. Low eV EI for compounds with abundant or dominant molecular ions in their 70 eV mass spectra results in the reduction of low mass fragment ions abundances thereby reducing their NIST library identification probabilities thus rarely justifies its use in real‐world applications.
3. Cold EI significantly enhances the relative abundance of the molecular ions particularly for large compounds; yet, it retains the low mass fragment ions; hence, Cold EI mass spectra can be effectively identified by the NIST library.
4. Different standard EI ion sources provide different 70 eV EI mass spectra. Among the Agilent technologies ion sources, the “Extractor” exhibits relatively abundant molecular ions compared with the “Inert” ion source, while the “High efficiency source” (HES) provides mass spectra with depleted molecular ions compared with the “Inert” ion source or NIST library mass spectra.

These conclusions are demonstrated and supported by experimental data in nine figures and two tables.     

The modern mass spectrometer—a complete chemical laboratory

Using nitrobenzene as an example, various ways in which a contemporary mass spectrometer can be utilized to yield a wealth of information about the compound studied are reviewed. Applying a variety of different techniques and procedures, in addition to the conventional low resolution mass spectrum, the following nitrobenzene spectra have been obtained: collision induced dissociation mass spectrum, mass analysed ion kinetic energy spectra, collision induced dissociation mass analysed ion kinetic energy spectra, spectra obtained at constant B/E, spectra obtained at constant B²/E, high voltage scans of metastable ion fragmentation processes, consecutive fragmentations in different field free regions, charge exchange mass spectra, charge stripping mass spectra, doubly charged ion mass spectra, chemical ionization mass spectra, negative ion mass spectra, negative ion mass analysed ion kinetic energy spectra, negative ion mass analysed ion kinetic energy collision induced dissociation spectra, charge inversion spectra, etc. The complementary types of information available from the above studies are discussed to show the unique versatility of mass spectrometry as a technique for the examination of organic compounds.     

Comparing the performance of neural networks to well-established methods of multivariate data analysis: the classification of mass spectral data

Summary The performance of neural networks in classifying mass spectral data is evaluated and compared to methods of multivariate data analysis and pattern recognition. Back propagation networks are matched with linear discriminant analysis, Kohonen feature maps are compared to the knearest neighbour clustering algorithm. Eight classifiers were trained, in order to discriminate mass spectra of steroids from eight distinct classes of chemical compounds. The results obtained show slightly better performance of Kohonen networks compared to k-nearest neighbour clustering and equal performance of multi-layer perceptrons and discriminant analysis.     

Comparative chemical fingerprinting of Oroxylum indicum and Scutellaria baicalensis using liquid chromatography and mass spectrometry

Introduction: Identification of Oroxylum indicum and Scutellaria baicalensis provides an interesting challenge in selection of biomarker compound to be used in routine analysis. Both plants have similar phytochemical profile and are rich sources of flavones and flavone glycosides. The objective of this study was to prepare the chemical fingerprinting of O. indicum bark and S. baicalensis roots using the liquid chromatography and mass spectroscopy in single chromatographic method. Materials and methods: Extracts prepared using various solvent systems (methanol, aqueous methanol, chloroform, hexane, and water) of both plants were analyzed using C18 reverse phase column with solvent system containing acetonitrile and 0.1% formic acid. Major flavonoids were identified based on mass spectra, fragmentation pattern, and UV spectra. Results: In this article, well-resolved high-performance liquid chromatographic (HPLC) separation in both plant extracts was obtained and chemical fingerprints for both plant extracts were established and flavonoids present (baicalin, oroxylin A-7-O-glucuronide, chrysin-7-O-glucuronide, baicalein, chrysin, oroxylin-A, wogonin, skullcap flavone II) were identified as possible biomarkers. Conclusion: Mass spectrometry coupled with HPLC can be a tool for fingerprinting of various natural products used in dietary supplement industry. The fingerprint developed in the article can be used for quality evaluation as well as identifying possible adulteration of extracts of both the plants.     

Analysis of Rotenoids by High Temperature High Resolution Gas Chromatography-Mass Spectrometry

A short glass capillary column coated with PS-090 (20% phenyl–80% methylpolysiloxane, of 10 m × 0.30 mm i.d.; with 0.1 μm film thickness) was used to analyze samples from Tephrosia candida; eighteen rotenoids were separated and identified without derivatization by HT-HRGC and HT-HRGC-MS. The mass spectra show the typical fragmentation pattern for rotenoids, with the base peaks either at M⁺, or originating from RDA (Retro Diels-Alder) rearrangements. HT-HRGC and HT-HRGC-MS were shown to be extremely valuable and neglected techniques for structural studies as well as the routine analysis of rotenoids in crude extracts. Possible applications to other classes of aromatic natural products, e.g. flavonoids, are envisaged.     

Schiff base derivatives of peptide esters: Relative abundance of N-terminal,C-terminal and ‘internal’ fragments as a function of the blocking group

Electron-impact (EI) mass spectrometry of peptide derivatives is usually interpreted in terms of fragmentation where the charge resides on the N-terminal fragments and to a lesser degree on the less common, charged C-terminal fragments. Substituted and unsubstituted benzylidene, cinnamylidene, α- and β-naphthylidene derivatives of a reference tripeptide, valileala, gave both N- and C-terminal fragments as well as molecular ions. The order of increasing ion current (normalized) in C-terminal fragments was: acetylacetonyl, 4-dimethylaminonaphthylidene, p-dimethyl-aminobenzylidene, 3-pyridylmethylidene, p-diethylaminocinnamylidene, benzylidene, 2-hydroxy-naphthylidene, 4-pyridylmethylidene, p-nitrobenzylidene, p-methoxybenzylidene, p-cyanobenzylidene, cinnamylidene, p-dimethylaminocinnamylidene, β-indolylmethylidene, β-naphthylidene, 2-pyridylmethylidene and α-naphthylidene. The order for this value among the N-terminal fragments is significantly different, however (Day, Falter, Lehman and Hamilton, J. Org. Chem. in press). In addition to N- and C-terminal fragments, many spectra contain internal fragments, arising from loss of fragments from both ends, which provide sequence information. These fragments are found in the mass spectra of Schiff bases formed from various aromatic aldehydes with peptide esters. The interpretation of the latter pattern is facilitate in some cases by deuterium labeling at the α-carbon of the N-terminal amino acid residue of peptides. Such a pattern provides sequence information supplemental to that available involving N- and C-terminal fragmentations. In derivatives of hexaglycine, tetraphenylalanine and tryptophylmethionylaspartyl (β-OEt) phenylalanine amide, for example, substantial sequence information was contained in the internal fragments; in some cases the sequence could be deduced only if the internal fragments were utilized. The 4-dimethylamino-naphthylidene derivatives have proven to be the most useful to date in terms of volatility, tendency to maximize cleavage into N-terminal fragments, intensity of molecular ions and generation of useful mass spectra of certain peptide esters refractory to mass spectrometry in the form of any other derivative investigated.     

The mass spectra and ionization potentials of the neutral fragments produced during the electron bombardment of aromatic compounds

A mass spectrometer equipped with a dual ionization chamber ion source has been used to characterize directly the neutral species produced in the dissociative ionization of gases by electron impact. Neutral fragment mass spectra have been obtained for the electron ionization and fragmentation of benzene, toluene, o-xylene, m-xylene, p-xylene, mesitylene and isotopically labeled toluene. The neutral fragment mass spectra correlate well with the structures of the molecules. The abundant species in the neutral fragment mass spectra also correlate reasonably well with the abundant complementary positive ions of the normal mass spectra. Ionization potentials have been determined for the abundant neutral species produced. Where comparisons with values reported elsewhere are possible, the agreement is usually within ±0.2 eV or less.     

Development of a new electron ionization/field ionization ion source for gas chromatography/time‐of‐flight mass spectrometry

We have developed a combined EI/FI source for gas chromatography/orthogonal acceleration time‐of‐flight mass spectrometry (GC/oaTOFMS). In general, EI (electron ionization) and FI (field ionization) mass spectra are complementary: the EI mass spectrum contains information about fragment ions, while the FI mass spectrum contains information about molecular ions. Thus, the comparative study of EI and FI mass spectra is useful for GC/MS analyses. Unlike the conventional ion sources for FI and EI measurements, the newly developed source can be used for both measurements without breaking the ion source vacuum or changing the ion source. Therefore, the combined EI/FI source is more preferable than the conventional EI or FI ion source from the viewpoint of the reliability of measurements and facility of operation. Using the combined EI/FI source, the complementarity between EI and FI mass spectra is demonstrated experimentally with n‐hexadecane (100 pg): characteristic fragment ions for the n‐alkane such as m/z 43, 57, 71, and 85 are obtained in the EI mass spectrum, while only the parent peak of m/z 226 (M⁺) without any fragment ions is observed in the FI mass spectrum. Moreover, the field desorption (FD) measurement is also demonstrated with poly(ethylene glycol)s M600 (10 ng) and M1000 (15 ng). Signals of [M+H]⁺, [M+Na]⁺ and [M+K]⁺ are clearly detected in the FD mass spectra. Copyright © 2009 John Wiley & Sons, Ltd.     

Merits of random forests emerge in evaluation of chemometric classifiers by external validation

Real-world applications will inevitably entail divergence between samples on which chemometric classifiers are trained and the unknowns requiring classification. This has long been recognized, but there is a shortage of empirical studies on which classifiers perform best in ‘external validation’ (EV), where the unknown samples are subject to sources of variation relative to the population used to train the classifier. Survey of 286 classification studies in analytical chemistry found only 6.6% that stated elements of variance between training and test samples. Instead, most tested classifiers using hold-outs or resampling (usually cross-validation) from the same population used in training. The present study evaluated a wide range of classifiers on NMR and mass spectra of plant and food materials, from four projects with different data properties (e.g., different numbers and prevalence of classes) and classification objectives. Use of cross-validation was found to be optimistic relative to EV on samples of different provenance to the training set (e.g., different genotypes, different growth conditions, different seasons of crop harvest). For classifier evaluations across the diverse tasks, we used ranks-based non-parametric comparisons, and permutation-based significance tests. Although latent variable methods (e.g., PLSDA) were used in 64% of the surveyed papers, they were among the less successful classifiers in EV, and orthogonal signal correction was counterproductive. Instead, the best EV performances were obtained with machine learning schemes that coped with the high dimensionality (914–1898 features). Random forests confirmed their resilience to high dimensionality, as best overall performers on the full data, despite being used in only 4.5% of the surveyed papers. Most other machine learning classifiers were improved by a feature selection filter (ReliefF), but still did not out-perform random forests.