The Aryl hydrocarbon receptor (AhR) is a ligand-dependent transcription factor. It regulates the expression of an array of genes in a wide range of species and tissues. Among the most characterized chemical classes that are known to be ligands for AhR are environmental toxins, such as the Halogenated Aromatic Hydrocarbons (HAHs) and Polycyclic aromatic hydrocarbons (PAHs).We utilized a large highly unbalanced dataset from PubChem Bioassay database (aid: 2796) of 324751 tested compounds with an active/inactive ratio of 1:40.  To evaluate the structural characteristics of small molecules responsible for AHR binding and toxicity, cross-validated classification models were developed using the online tool OCHEM (http://ochem.eu). Multiple machine learning algorithms including ASNN, KNN, FSMLR, PLS, Random forests and J48 were compared. The descriptors packages used for modeling the dataset included Dragon v6, CDK, ALOGPS, ESTATE, ISIDA, ADRIANA.CODE as well as Chemaxon descriptors. Stratified Bootstrap aggregation was used to successfully handle the unbalance in the dataset. Distance-to-model applicability domain was estimated. Enrichments of 3-4 folds were observed for the most accurate predictions. Structural features contributing to the activation of AhR receptor were suggested. The most accurate models are made freely available online on http://ochem.eu.

Development of classification models for the Aryl Hydrocarbon Receptor (AHR) activators using HTS big dataset: Utilizing applicability domain and addressing unbalance in the dataset

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.