Consensus ranking approach to understanding the underlying mechanism with QSAR

Area of Science:

• * Cheminformatics and computational toxicology.
• * Quantitative structure-activity relationship (QSAR) studies.

Background:

• * QSAR aims to build predictive models and elucidate chemical mechanisms.
• * Mechanistic analysis in QSAR is often secondary to model construction.
• * Conventional descriptor selection methods are sensitive to training set composition.

Purpose of the Study:

• * To evaluate descriptor selection methods for QSAR mechanistic analysis.
• * To develop a robust protocol for selecting descriptors for mechanistic insights.
• * To compare the performance of descriptors selected for prediction versus mechanism.

Main Methods:

• * Proposed a consensus ranking protocol for descriptor selection.
• * Evaluated descriptor robustness against variations in training chemical sets.
• * Compared model performance using descriptors optimized for prediction versus mechanism.

Main Results:

• * Conventional descriptor selection methods are unsuitable for mechanistic analysis.
• * The consensus ranking protocol yields robust descriptors for mechanistic analysis.
• * Descriptors selected for mechanistic analysis resulted in inferior predictive model performance.

Conclusions:

• * Mechanistic analysis in QSAR requires distinct descriptor selection strategies.
• * A consensus ranking protocol offers a robust approach for mechanistic descriptor selection.
• * Dedicated model development is crucial for achieving optimal predictive capability in QSAR.