Ecotoxicity prediction using mechanism- and non-mechanism-based QSARs: a preliminary study

Area of Science:

• Ecotoxicology
• Computational toxicology
• Structure-activity relationship modeling

Background:

• Mechanism-based quantitative structure-activity relationships (QSARs) often yield higher quality predictions than non-mechanism-based QSARs.
• Accurate determination of a compound's toxicity mechanism is crucial but challenging for mechanism-based QSARs.
• Potential errors in mechanism identification can compromise the predictive performance of mechanism-based QSARs.

Purpose of the Study:

• To compare the efficacy of the mechanism identification-toxicity prediction (MI-TP) approach against the direct toxicity prediction (DTP) approach.
• To evaluate the impact of mechanism prediction errors on the overall accuracy of toxicity predictions.
• To assess the performance of mechanism-based QSARs versus non-mechanism-based QSARs in predicting phenol toxicity.

Main Methods:

• Developed a statistical model for mechanism classification to predict compound mechanisms.
• Constructed four mechanism-based QSARs and one QSAR disregarding mechanism.
• Applied both MI-TP and DTP approaches to predict the toxicity of phenols in an external dataset using Tetrahymena pyriformis.

Main Results:

• Incorrect mechanism predictions for several phenols in the external test set led to significant over- or under-estimation of their toxicity.
• The MI-TP approach did not demonstrate superior accuracy compared to the DTP approach in predicting phenol toxicity.
• The inclusion of mechanism identification did not consistently enhance the predictive performance of QSAR models.

Conclusions:

• The mechanism identification step in the MI-TP approach can introduce errors that negate potential benefits.
• Direct toxicity prediction (DTP) may be as effective, or more effective, than MI-TP for toxicity prediction in certain contexts.
• Further research is needed to refine mechanism identification methods or explore alternative QSAR strategies for improved ecotoxicity predictions.