Predicting carcinogenicity by using batteries of dependent short-term tests

Area of Science:

• Toxicology
• Biostatistics
• Computational Biology

Background:

• The carcinogenicity prediction and battery selection (CPBS) procedure is a prominent method for predicting chemical carcinogenicity using short-term tests (STTs).
• A key assumption of CPBS is the conditional independence of the STTs employed.
• Recent National Toxicology Program (NTP) studies have challenged this independence assumption for commonly used in vitro STTs.

Purpose of the Study:

• To address the violation of the conditional independence assumption in the CPBS procedure.
• To propose a modified approach for carcinogenicity prediction that accounts for dependencies among STTs.
• To enhance the accuracy and reliability of carcinogenicity predictions derived from STT batteries.

Main Methods:

• Modification of the existing CPBS procedure.
• Application of log-linear modeling to accommodate dependencies between STTs.
• Analysis of data from National Toxicology Program (NTP) studies involving in vitro STTs.

Main Results:

• The conditional independence assumption of STTs in CPBS was found to be violated by recent NTP study results.
• Log-linear modeling was successfully employed to modify CPBS, accounting for inter-test dependencies.
• The modified approach provides standard errors for predicted probabilities of carcinogenicity, improving uncertainty quantification.

Conclusions:

• The standard CPBS procedure requires modification due to demonstrated dependencies among STTs.
• Log-linear modeling offers a robust method for adjusting CPBS to account for these dependencies.
• The revised approach enhances the predictive power and statistical rigor of carcinogenicity assessments using STT batteries.