Spectrum interpretation library search

INFERCNMR, as the interpretive library searches described earlier, is basically a subspectrum-matching procedure. The program first retrieves all subspectra from the reference library, at least six signals of which (default value) match different signals in the query spectrum which are of the same multiplicity and within a user-set chemical shift tolerance. If the signals of a retrieved reference library subspectrum correspond to at least six connected carbon atoms (default value), that reference library substructure is predicted to be present in the compound of the unknown structure. 

The similarity search and the interpretive library search, although both classed as library searches, are actually quite different. The former retrieves spectra whose overall properties are similar to the spectrum of the unknown, The corresponding structures reveal overall structural similarity, although common substructures may be extracted. In contrast, the interpretive library search only requires a subspectrum match overall structural similarity between a reference compound and the unknown is not necessary in order to retrieve a substructure common to both. 

The neural networks described share some limitations with library search procedures. They are effective in interpolating, but not in extrapolating. Therefore, their utility in spectrum interpretation is limited by the diversity of the database used in training. Although both the neural network and the interpretive library search are database dependent, the former requires a predefined set of structural features the latter need not. 

Third, assuming that the required information can be automatically gathered, it will be necessary to use an expert system approach to automate the data interpretation. Such a system will likely employ both library searching and IR and MS spectrum interpreters to analyze the data, making use of intermediate results from one approach to guide the others in the same way that a human operator interprets all of the data in concert. Combined library search approaches have been demonstrated (16,17,T9) and a variety of MS (20-22) and IR (23-25) spectrum interpreters are available. Several laboratories have begun to address the need for a combined expert system (16,22). It seems clear that some of the most useful new developments in GC/IR/MS technology will appear in this area. 

The spectrum is examined first. Before starting any interpretation, it is strongly recommended that a computer or a manual library search is performed to check whether this spectrum belongs to an existing collection. Identification of an unknown compound in this way depends directly on the quality and comprehensiveness of the collection used. However, only libraries of electron ionization spectra are efficient. Other ionization techniques yield spectra that are much too dependent on the instruments and experimental conditions. 

For qualitative spectrum interpretation, the conventional method for routine identification of chemical species is a library-search, based on spectral mapping algorithms. Before library-searching spectral preprocessing, i.e., elimination of baseline effects and noise, standardization, etc., is performed on the sample spectrum. Comparison of such a processed spectrum with a... 

Some differences are also seen in the El spectra for the same type of compound but generated from a different initial monosaccharide. However, these differences are sometimes more difficult to interpret, as the identifications rely on mass spectral library searches. The spectrum for tri-TMS 1,6-anhydroglucopyranose was shown in Figure... 

The mass library searches are in fact elaborate electronic matching routines, which provide at the end a list of possible answers and for each of them a calculated percentage match. The rules for explaining the mass spectrum of a particular compound that were developed to interpret mass spectra without the help of library searches are now more useful for choosing the correct structure from a list of possible compounds identified as candidates by the library search or for obtaining molecular information when the library search does not give a proper answer. 

So spectrum interpretation or library searching may be frustrated unless SERS reference spectra are available. Fourth, a practical substrate for routine analysis should be available commercially and have a long shelf life. At this writing, all SERS substrates are prepared by individual labs, usually immediately before use. The world needs a stable, reproducible, and probably disposable SERS substrate in commercial quantities, but so far none of the various substrates... 

A representative spectrum of each peak in turn is examined, after background subtraction if necessary, and submitted to a library search. An experienced GC-MS analyst must interpret the validity of the library matches. If the library match is considered to be unequivocal, the identification is reported. If isomer specificity is not certain, but elemental composition is considered unequivocal the fit is reported without specific isomer identifiers. If the compound type is unequivocal, but specific compound identification is not possible, then the compound class is reported. If no library matches are considered to be reasonable, then the component is designated as unknown. Individual laboratories should always establish their own criteria for reporting compounds as positively identified. 

While a spectrum interpretation can be a very involved and tedious task while dealing with a new structure, many spectra need not be interpreted from the first principles . Once a spectrum has been recorded, it is now soon included into one of the growing libraries of mass-spectral information. Many thousands of mass spectra from biological and environmentally important substances have been acquired. A computer search for such compounds is a relatively straightforward task. A mass spectrum from a particular sample can be compared to the reference spectra that have been accumulated over the years. The individuals specializing in certain compound types may also have their own mass-spectral libraries. 

Two methods are commonly applied for library searches. Identity or retrieval searches assume that the spectrum of the unknown compound is present in the reference library, and only experimental variability prevents a perfect match of the unknown and reference spectra. When no similar spectra are retrieved the only information provided is that the unknown spectrum is not in the library. Similarity or interpretive searches assume that the reference library does not contain a spectrum of the unknown compound, and are designed to produce structural information from which identity might be inferred. Interpretive methods typically employ a predetermined set of spectral features, designed to correlate with the presence of chemical substructures. Searching identifies the library spectra that have features most similar to those of the unknown spectrum. The frequency of occurrence of a substructure in the hit list is used to estimate the probability that it is present in the unknown compound. Two well-developed interpretative search algorithms are SISCOM (Search of Identical and Similar Compounds) and STIRS (the Self-Training Interpretive and Retrieval System) [174-177]. Normally a retrieval search is performed first, and when the results are inconclusive, an interpretive search is implemented. In both cases, success depends on the availability of comprehensive libraries of high-quality reference spectra [178]. 

The interpretation of spectra is considerably facilitated by modem data processing. In commercial libraries, e.g. the NIST 75.000 or the Pfleger-Maurer-Weber library for drugs and pesticides, reference spectra are available that can be compared with the measured spectra, the degree to which the measured spectrum matches the entry in the library being known as the similarity . The library search can be fully automated (Fig. 4-6), which represents a considerable saving in time. 

A KNN-classification is almost identical with the interpretation of spectra by a library search. In library search an unknown spectrum (pattern) is compared with all spectra of known compounds collected in a spectral library. A similarity criterion or a dissimilarity criterion (equivalent to a distance measurement) between two spectra must be defined. To find the most similar spectra in the library, this criterion must be calculated for each library spectrum. 

Three different approaches have been used for computer-assisted interpretations of chemical data. 1. Heuristic methods try to formulate computer programs working in a similar way as a chemist would solve the problem. 2. Retrieval methods have been successfully used for library search (an unknown spectrum is compared with a spectral library). 3. Pattern recognition methods are especially useful for the classification of objects (substances, materials) into discrete classes on the basis of measured features. A set of characteristic features (e.g. a spectrum) of an object is considered as an abstract pattern that contains information about a not directly measurable property (e.g. molecular structure or biological activity) of the object. Pure pattern recognition methods try to find relationships between the pattern and the "obscure property" without using chemical knowledge or chemical prejudices. 

Spectrum interpretation supported by automatic library search (ionisation mode dependent), spectra subtraction, etc. 

Spectrum interpretation supported by automatic library search... 

The library search will possibly only lead to a correct identification if the spectrum of the unknown is actually present in the library and the GC separation has been sufficiently efficient to obtain a sufficiently clean mass spectrum. When the unknown is not present in the library, the search procedure also yields valuable information in pointing out certain structural elements present in the unknown as well as structural similarities with known compounds. However, this information is only useful in combination with a proper interpretation of the mass spectrum. A fast check of the mass spectra found by the search routine against the supposed structure is advisable as well, in order to eliminate possible errors in the library. The Wiley Library, for example, contains almost identical mass spectra for 2-butanol and 2-methyl-1-propanol, while theoretically and practically, these mass spectra are distinctly different. It must also be taken into account that the vast majority of the spectra available in the library are of compounds having molecular masses between 150 and 300 Da. The number of spectra of compounds with molecular masses above 400 Da is limited, although the number... 

Next to the library search, the interpretation of mass spectra is an important aspect of identification of unknowns. A huge amount of information on the fragmentation reactions of molecular ions and the resulting fragments is available [23]. Helped by the knowledge of possible functionalities and directions from the library search, the identity of an unknown may be determined by interpretation of the spectrum. Sometimes, this can be quite difficult. The description of the various fragmentation reactions is beyond the scope of this book. 

Mass spectra of chemical compounds have a high information content. This article describes computer-assisted methods for extracting information about chemical structures from low-resolution mass spectra. Comparison of the measured spectrum with the spectra of a database (library search) is the most used approach for the identification of unknowns. Different similarity criteria of mass spectra as well as strategies for the evaluation of hitlists are discussed. Mass spectra interpretation based on characteristic peaks (key ions) is critically reported. The method of mass spectra classification (recognition of substructures) has interesting capabilities for a systematic structure elucidation. This article is restricted to electron impact mass spectra of organic compounds and focuses on methods rather than on currently available software products or databases. 

Figure 7 Example of a library search by MassLib. A mass spectrum from testosterone has been considered as unknown and searched in a library consisting of 130 000 spectra (including duplicates). A mass spectrum from testosterone is contained in this library and has been found as the first hit (most similar spectrum). The other hits are from the same compound class and demonstrate the interpretive capabilities of this system.

Most commercial FTIR software packages come equipped with a library searching capability, or it can be purchased for an additional fee. When a spectral library search is performed, the unknown spectrum is compared to each spectrum in the libraries selected. Thus, if there are 1000 spectra in a library, 1000 comparisons are performed. As a result of each comparison a number called the hit quality index (HQI) is calculated, which is a numerical measure of the similarity between two spectra. In an ideal world the library search will turn up a spectrum similar to the unknown spectrum. It is then assumed the unknown sample is chemically similar to the library sample. This way known spectra can be used to identify unknown spectra. Library searching is so useful in identifying unknowns and interpreting mixture spectra that I encourage all FTIR users to have access to library searching capabilities. 

---