Unknowns chemical classification

Antibiotics have a wide diversity of chemical stmctures and range ia molecular weight from neat 100 to over 13,000. Most of the antibiotics fall iato broad stmcture families. Because of the wide diversity and complexity of chemical stmctures, a chemical classification scheme for all antibiotics has been difficult. The most comprehensive scheme may be found ia reference 12. Another method of classifyiag antibiotics is by mechanism of action (5). However, the modes of action of many antibiotics are stiU unknown and some have mixed modes of action. Usually within a stmcture family, the general mechanism of action is the same. For example, of the 3-lactams having antibacterial activity, all appear to inhibit bacterial cell wall biosynthesis. 

The following chapter comprises those dyestuffs which cannot be placed under the foregoing chemical classifications. It has already been mentioned that many of the dyestuffs treated of under the previous groups are of unknown constitution but, being products of chemical synthesis, certain relationships to the various classes are easily deduced. 

Analysis of Nontarget Compounds. "Complete Unknowns. This is a somewhat similar process in that the retention time and the type of LC column giving the best results also yields dues as to chemical classification, e.g., good retention and separation upon an anion exchange column suggests that the analytes are anionic. Confirmation information required for unknown identification is also obtained from other instrumentation induding UV spectrophotometry. 

The importance of validation has been generally acknowledged, and most QSAR models in the literature are validated either by cross validation or external test sets [13,46]. Model validation for classification models is typically specified by statistical quality measures of overall quality such as sensitivity, specificity, false positives, false negatives, and overall prediction. Unfortunately, it is often impossible to specify accuracy and prediction confidence for individual unknown chemicals, specifically those unknown chemicals with structures requiring the model to extend to, or beyond, the limits of chemistry space defined by the training set. 

Figure 6.4 Illustration of the decision forest (DF). The individual trees are developed sequentially, and each tree uses a distinct set of descriptors. Classification (i.e., prediction) of an unknown chemical is based on the mean results of all trees.

Figure 6.9 The focused domain representing the training domain of a tree. For an unknown chemical predicted by the tree, its classification is determined by a terminal node (e.g., dark circle) to which it belongs. There are three descriptors used in the path (bold line) from the root to the terminal node and the range of these three descriptors across all chemicals in the training set determines the training domain.

Yet, when one considers the number of chemicals which is perhaps in excess of 10 million, whereas the largest of the databases of mass spectra contain some 200000 chemicals, the need to employ other techniques is obvious. In this way, and coexisting with the development of databases some software packages use the concept of artificial intelligence methods to determine chemical structures from their mass spectra. In a number of cases these techniques based upon programmable criteria, chemical classification of the unknown, rarity of mass etc. appear to work well, but are limited to particular classes of compounds. The problem lies in the fact that unknowns are more often totally unknown (such as in environmental samples). 

Coal Tar Products. Evaluations of human exposure during employment in coal tar plants indicated significant increases in cancer-related deaths (TOMA 1982). No specific type of cancer was predominant. Nevertheless, no clear relationship could be established because exposure routes in addition to dermal were likely, such as inhalation and oral. Also, the ability to relate death to coal tar exposure was further confounded by the possibility that the subjects were also exposed to other chemicals and cigarette smoke (TOMA 1982). Additional limitations were identified in this study, including absence of data on smoking habits, short cut-off date of 10 days of employment, unknown race classification for 20 of participants, use of U.S. male mortality rates for comparison as opposed to regional mortality rates, and the relationship between the cohort and production history was not explored. 

The standard transinformation R is not necessarily the maximum amount of information that can be obtained by a given classifier. The maximum amount R is called channel capacity C417, 4273. R would be obtained for a random sample of patterns with optimally adjusted a priori probabilities. Because the optimal a priori probabilities depend on the classifier, the channel capacity is not suitable for the evaluation of classifiers. The only useful starting point for the classification of unknown chemical patterns is to assume equal probabilities for all classes. 

As long as our familiarity with organic bodies was merely superficial and limited itself primarily to external properties, while transformations and mutual relations remained almost entirely unknown, the classifications of organic compounds, considered firom the standpoint of chemistry, could not be natural. At that time, bodies were grouped according to their origin, color, consistency, etc., divided into volatile oils, resins, dyestuffs, etc. A closer, but not yet complete famdiarity with the chemical properties of a part of the organic compounds led to the distinction between acids, alkalis, and indifferent bodies. 

Because classification is a supervised problem the classes should be known and assigned to the objects from the training set in advance. No matter which ANN techniques is used the targets should be represented as binary vectors having as many components as there are prespecified classes. Therefore, the variables have only two values zeros and ones. If one deals with a one-object-to-one-class problem only one target component at one time is allowed to be different from zero. On the other hand, when one object can be associated with several classes simultaneously, more components of the target can differ from zero. The latter case is typical for spectra-structure correlation problems where an unknown chemical compound, represented as a spectrum (IR, mass, CNMR, UV, etc.) is associated with several targets, i.e., with several... 

Exploratory data analysis shows the aptitude of an ensemble of chemical sensors to be utilized for a given application, leaving to the supervised classification the task of building a model to be used to predict the class membership of unknown samples. 

Concluding this section, we may mention a paper by Daams and Villars (1993) concerning an atomic environment classification of the chemical elements. Critically evaluated crystallographic data for all element modifications (and recommended atomic volumes) have been reported. Special structural stability diagrams were used to separate AET stability domains and to predict the structure (in terms of environment types) of hitherto unknown high-pressure and high-temperature modifications. Reference to the use of AET in thermodynamic (CALPHAD) modelling and calculation has been made by Ferro and Cacciamani (2002). 

As a result of machine learning a model is produced of the characteristic exhibition of a property (for instance, the formation of a particular type of chemical compound) which corresponds to a distribution pattern of this property in the multidimensional representative space of the properties of the elements. The subsequent pattern recognition corresponds to a criterion for the classification of the known compounds and for the prediction of those still unknown. Examples of this approach reported by Savitskii are the prediction of the formation of Laves phases, of CaCu5 type phases, of compounds XY2Z4 (X, Y any of the elements, Z = O, S, Se, Te), etc. (Data on the electronic structures of the components were selected as... 

The industrial stream to be treated, the feed, will not be an analytical grade solute dissolved in water, but often contains several known and unknown substances, both organic and inorganic. To be able to make an initial selection of possible solvents, it is necessary to make a classification of the individual substances present in the feed and of the groups of substances with chemical similarities, for instance, paraffins, aromatics, salts, or others. The following questions have to be answered ... 

Pattern recognition can be applied for the determination of structural features of unknown (monofunctional) compounds (Huber and Reich ). The information about the chemical structure is contained in a multidimensional gas-liquid retention data/stationary liquid phases set. The linear learning machine method is applied in a two step classification procedure. After the determination of a correction term, the skeleton number, a classification step for the determination of the functional group is executed. It is remarkable that 10 stationary phases are sufficient for the classification. 

The example of Fig. 3 shows the use of blind chemical analysis, where the peaks of a chromatogram do not require chemical interpretation they are simply the fingerprint of the objects. These patterns allow the prediction of the category of an unknown object by its position in the plot, without using any classification method. 

Lipids, relatively nonpolar chemical substances found in plant, bacterial, and animal cells, are among the most ubiquitous of biomolecules. In this experiment, a lipid extract of ground nutmeg will be purified by chromatography on a silica gel column. Analysis of the lipid extract by thin-layer chromatography will provide the classification of the components in the extract. The unknown lipids will be further characterized by saponification and analysis of the fatty acid content by gas chromatography. For an abbreviated experiment, students may be provided samples of natural oils and fats that can be analyzed by saponification and gas chromatography. 

The classification tests (summarized in Table 31.2), when properly done, can distinguish between various types of aldehydes and ketones. However, these tests alone may not allow for the identification of a specific unknown aldehyde or ketone. A way to correctly identify an unknown compound is by using a known chemical reaction to convert it into another compound that is known. The new compound is referred to as a derivative. Then, by comparing the physical properties of the unknown and the derivative to the physical properties of known compounds listed in a table, an identification can be made. 

Fig. 3.2 Typical applications using chemical multivariate data (schematically shown for 2-dimensional data) cluster analysis (a) separation of categories (b), discrimination by a decision plane and classification of unknowns (c) modelling categories and principal component analysis (d), feature selection (X2 is not relevant for category separation) (eY relationship between a continuous property Y and the features Xi and X2 (f)...

The elucidation of the primary site and biochemical mode of action of inhibitors is often difficult and not necessarily associated with the mechanism of resistance in field isolates. Nevertheless, information on the mechanism of resistance can provide evidence to determine the site of action. Therefore, the classification of fungicides is based on crossresistance reactions rather than chemical similarities of structures or proposed modes of action (Table 1). Based on available information in the literature, three categories of inhibitor classes can be made Classes with known mode of action and known mechanism of resistance, classes with proposed mode of action and unknown mode of resistance but wide-spread field resistance, and classes in which resistance is claimed to occur in the field but both mode of action and resistance are not known. 

---