A vast amount of chemical toxicology and property data is publicly accessible via the Internet. However, these data are often uncurated, unreferenced, distributed across many data sources, and can contain a myriad of data quality issues. This project sought to develop a systematic approach to consolidate existing chemical data for use in quantitative structure-activity relationship (QSAR) modeling. A large compilation of physicochemical data (&gt;2 million data points) was collected from 16 publicly available sources using automated tools built in Java. These data were converted to a consistent machine-readable format, and stored in an SQLite database. The use of SQL queries allowed for the convenient assembly of data subsets by characteristics such as experimental property, conditions, and test methods. The experimental data were filtered for QSAR validity (e.g. eliminating implausible property values and constraining experimental ambient conditions), and substances were mapped to unique substance identifiers (DTXSIDs) using the EPA’s Distributed Structure-Searchable Toxicology (DSSTox) Database to obtain structural data. The structural data were filtered again for QSAR validity (e.g. removing salts and metallic atoms) and used to generate molecular descriptor values. Finally, records were stored in “QSAR-ready” form (i.e. desalted non-stereoforms with isotopes removed) for use as input to a variety of existing and newly developed QSAR models. The development of automated data collection tools, as well as web services called via an application programming interface (API) for individual steps of data preparation and modeling, created a generalizable workflow. This workflow could be applied to any type of experimental data; for any set or subset of substances of interest; with any desired constraints, descriptors, and methods for modeling. The effectiveness and generality of this system was demonstrated through data gathering and modeling efforts on water solubility, skin sensitization, skin irritation, and eye irritation.

Systematic development of QSAR data sets from online data

ACS Fall 2021

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