Constructing a Structure-Based Chemical Ontology from Existing Nomenclature

Each node or category name in ClassyFire’s chemical ontology or ChemOnt, was created by extracting common or existing chemical classification category terms from the scientific literature and available chemical databases. We used existing terms to avoid “reinventing the wheel”. By making use of commonly recognized or widely used terms that already exist in the chemical literature, we believed that the taxonomy (and the corresponding ontology) should be more readily adopted and understood. This dictionary creation process was iterative and required the manual review of a large number of specialized chemical databases, textbooks and chemical repositories. Because the same compounds can often be classified into multiple categories, an analysis of the specificity of each categorical term was performed. Those terms that were determined to be clearly generic (e.g. organic acid, organoheterocyclic compound) or described large numbers of known compounds were assigned to SuperClasses. Terms that were highly specific (e.g. alpha-imino acid or derivatives, yohimbine alkaloids) or which described smaller numbers of compounds that clearly fell within a larger SuperClass were assigned to Classes or SubClasses. This assignment also depended on their relationship to higher-level categories. In some cases multiple, equivalent terms were used to describe the same compounds or categories (imidazolines vs. dihydroimidazoles). To resolve these disputes, the frequency with which the competing terms were used was objectively measured (using Google page statistics or literature count statistics). Those having the highest frequency would generally take precedence. However, attention was also paid to the scientific community and expert panels. When available, the IUPAC term was used to name a specific category. Otherwise, if the experts clearly recommended a set of (less frequently used) terms, these would take precedence over terms initially chosen by our initial “popularity” selection criteria. Examples include the terms “Imidazolines” (229,000 Google hits) and “Dihydroimidazoles” (4590 Google hits). The other popular terms were then added as synonyms. A total of 9012 English synonyms were added to the ChemOnt terminology data set.
In a number of cases, new SuperClass and Class terms were created for chemical categories not explicitly defined in the literature. Of these, the resulting “novel” categories were typically constructed from the IUPAC nomenclature for organic and inorganic compounds. Because our chemical dictionary was built from extant or common terms, it contains many community-specific categories commonly used in the (bio-)chemical nomenclature (e.g. primary amines, steroids, nucleosides). Moreover, due to the diverse nature of active and biologically interesting compounds, many chemical categories linked to specific chemical activities or based on biomimetic skeletons (e.g. alpha-sulfonopeptides, piperidinylpiperidines) were added. For instance, several compounds from the category of imidazo[1,2-a]pyrimidines (CHEMONTID:0004377) have been shown to display GABA(A) antagonist activity, and a potential to treat anxiety disorders [35 (link)].
After all the dictionary terms were identified and compiled (4825 terms to date), each term was formally defined using a precise, yet easily understood text description that included the structural features corresponding to that chemical category (Fig. 3). These formal definitions and the corresponding category mappings formed the basis of the structural classification algorithm and the classification rules described below. Once defined, the terms in this Chemical Classification Dictionary were progressively added to the taxonomic structure to form the structure-based hierarchy underlying ClassyFire’s chemical classification scheme. With the combination of the taxonomic structure and the Chemical Classification Dictionary, ChemOnt can be formally viewed as an ontology (albeit purely a structural ontology).Fig. 3

The chemical taxonomy. The taxonomy is illustrated with the OBO-Edit software, showing definitions synonyms, references, and extended information