Complementary PLS and KNN algorithms for improved 3D-QSDAR consensus modeling of AhR binding

Area of Science:

• Computational chemistry
• cheminformatics
• Toxicology

Background:

• Predicting aryl hydrocarbon receptor (AhR) binder activity is crucial for assessing toxicity.
• Quantitative structure-activity relationship (QSAR) models require robust validation and reliable descriptors.
• Existing models may not fully capture the complex relationships between chemical structure and biological activity.

Purpose of the Study:

• To develop a reliable consensus model for predicting the activity of aryl hydrocarbon receptor binders.
• To improve the predictive performance beyond individual modeling techniques.
• To utilize three-dimensional spectral data-activity relationship (3D-SDAR) descriptors for enhanced predictions.

Main Methods:

• A consensus partial least squares (PLS)-similarity based k-nearest neighbors (KNN) model was developed.
• The model utilized 3D-SDAR fingerprint descriptors for predicting log(1/EC50) values.
• Multiple validation techniques, including Y-scrambling and set randomization, were employed to ensure model reliability.

Main Results:

• The consensus model achieved a R2test of 0.685, showing a 10.5% improvement over individual models.
• The model was constructed from a PLS model (R2test=0.617) and a KNN model (R2test=0.622).
• The improvement was attributed to the complementary information from 3D-SDAR matrices of varying granularity.

Conclusions:

• The consensus PLS-KNN model demonstrates favorable predictive performance for Aryl hydrocarbon (AhR) binders compared to previous studies.
• The 3D-QSDAR approach provides valuable structural interpretation, aiding in the identification of toxicity-related features.
• The developed modeling strategy offers a reliable tool for predicting AhR binder activity and understanding structure-toxicity relationships.