Cluster analysis of short-term tests: a new methodological approach

Area of Science:

• Toxicology
• Computational Biology
• Data Science

Background:

• Evaluating the performance of short-term toxicity tests is crucial for predicting chemical hazards.
• Existing evaluation methods may not fully capture the nuances of test system responses.
• A data-driven approach can offer objective insights into test performance.

Purpose of the Study:

• To propose and validate a data-based methodology for evaluating short-term toxicity tests.
• To classify 22 different tests based on their performance across 42 chemicals.
• To identify patterns in test responses for improved hypothesis formulation and practical application.

Main Methods:

• Utilized literature data on 42 chemicals and the responses of 22 short-term tests.
• Applied cluster analysis with two different methods to classify the tests.
• Compared test performances based solely on their chemical response profiles.

Main Results:

• Cluster analysis consistently resolved the 22 tests into three distinct groups.
• Cluster 1 demonstrated high sensitivity but low specificity for carcinogen discrimination.
• Cluster 3 exhibited the opposite characteristics, while Cluster 2 showed intermediate performance.
• Test system specificity was a dominant factor in cluster membership, overriding phylogeny and endpoint considerations.

Conclusions:

• The proposed data-based approach effectively classifies short-term toxicity tests.
• Understanding the underlying patterns in test performance aids in selecting appropriate assays.
• This methodology provides a valuable tool for hypothesis generation and practical hazard assessment.