Predictive modeling of drug induced toxicology via QSAR analysis of SEND datasets

Rodney Moore Jr.

Mentor: Dr. Kevin Snyder U. S. Food and Drug Administration, Center for Drug Evaluation and Research, Office of New Drugs

Date/Time: August 23rd, 2022 at 3:00pm.

Abstract: According to the U.S. News, pharmaceutical products are one of the leading causes of death for Americans. The Food and Drug Administration (FDA) of the United States Department of Health is responsible for protecting the country through regulation of prescriptions and other pharmaceutical drugs, vaccines, etc. To promote user safety, the FDA has recently begun to require sponsor companies to submit standardized electronic study data along with toxicology study reports to support the safety of investigational pharmaceutical products.

Using the Global Substance Registration System (GSRS), the FDA’s online substance database, we will download nonclinical information tables and join these tables using SQL to produce a new dataset containing relevant attributes to describe and identify each instance of our data. The key attributes needed for the analyzation are approval id, SMILES (Simplified Molecular Input Line Entry System), application number, duration of drug dosage, method of drug taken, sex, species, and terminal body weight. With this dataset, we use Quantitative structure activity relationship (QSAR) modeling to qualify the safety of drug substances by robustly demonstrating toxicology endpoints.

Using chemmodlab, a cheminformatics modeling R package for fitting and assessing machine learning models, we created graphical user interfaces, to analyze study data and generate visualizations to submit alongside toxicology study reports. Terminal body weight was used as the sole predictor of identifying toxic chemicals. As determined by our findings, generally drugs that lead to a negative terminal body weight below a threshold of -1, were deemed to be more likely to be toxic and unsafe for consumers.