Generating Modeling Data From Repeat-Dose Toxicity Reports.

Area of Science:

• Toxicology
• Data Science
• Computational Chemistry

Background:

• Pharmaceutical companies have generated extensive in vivo repeat-dose toxicity (RDT) data for regulatory submissions over decades.
• The eTOX project has compiled numerous RDT studies into a centralized database, creating a valuable resource.
• High heterogeneity in experimental conditions across studies necessitates data normalization and standardization for analysis.

Purpose of the Study:

• To describe a method for extracting comparable toxicological datasets from heterogeneous RDT studies.
• To generate QSAR-like tables by mapping toxicological findings to specific in vivo endpoints.
• To demonstrate the utility of the processed data for developing predictive toxicology models.

Main Methods:

• Normalization of terminology used in original RDT reports.
• Filtering of studies based on experimental conditions to ensure comparability.
• Extraction of comparable toxicological profiles and mapping them to relevant in vivo endpoints.
• Generation of QSAR-like tables suitable for predictive modeling.

Main Results:

• A detailed strategy and associated tools for transforming raw RDT data into a usable format for modeling were developed.
• Proof-of-concept predictive models for in vivo liver endpoints were successfully built using the generated datasets.
• The processed data demonstrated suitability for developing hazard-predicting models.

Conclusions:

• The described method effectively transforms heterogeneous RDT data into standardized, comparable datasets.
• The generated QSAR-like tables are valuable for building predictive models of toxicological hazards.
• This approach facilitates the exploitation of existing RDT data for regulatory and safety assessment purposes.