Iterative fragment selection: a group contribution approach to predicting fish biotransformation half-lives

Area of Science:

• Environmental Chemistry
• Computational Toxicology
• cheminformatics

Background:

• Regulatory bodies require hazard and risk assessments for numerous chemicals with limited data.
• Existing methods for developing Quantitative Structure-Activity Relationships (QSARs) often rely on expert knowledge for fragment selection.

Purpose of the Study:

• To present an automated method for developing and evaluating QSARs applicable to screening-level chemical assessments.
• To generate novel QSARs for fish primary biotransformation half-lives (HL(N)) without requiring prior expert knowledge.

Main Methods:

• An integrated algorithm for descriptor generation, dataset splitting, cross-validation, and model selection was developed.
• The method produces two-dimensional (2D) fragment-based group contribution models.
• The approach incorporates domain of applicability and uncertainty estimates for predictions.

Main Results:

• Novel QSARs for HL(N) were generated and compared to an expert-developed model, showing similar predictive power.
• The automated method achieved a coefficient of determination (R(2)) of 0.789 and RMSE of 0.526 on training data, and R(2) of 0.748 and RMSE of 0.584 on validation data.
• The new QSAR models made comparable predictions using significantly fewer fragments than the expert-developed model.

Conclusions:

• The automated QSAR development method is effective for screening-level chemical assessments.
• This approach offers a valuable alternative to expert-driven QSAR development, particularly for properties like fish biotransformation half-life.
• The method provides reliable predictions with quantifiable uncertainty and domain of applicability.