Developing QSAR Models with Defined Applicability Domains on PPARγ Binding Affinity Using Large Data Sets and Machine...

Area of Science:

• Toxicology
• Computational Chemistry
• Drug Discovery

Background:

• Chemicals can disrupt human health by binding to peroxisome proliferator-activated receptor gamma (PPARγ).
• Assessing PPARγ binding affinity is vital for identifying potential endocrine-disrupting chemicals.
• Quantitative structure-activity relationship (QSAR) models with defined applicability domains (ADs) are essential for efficient screening, but are limited by data availability.

Purpose of the Study:

• To develop robust QSAR regression models for predicting PPARγ binding affinity.
• To introduce novel methods for defining applicability domains (ADs) and analyzing structure-activity landscapes (SALs).
• To enhance the reliability and interpretability of QSAR models for chemical safety assessment.

Main Methods:

• Curated PPARγ binding affinity data from diverse sources were utilized.
• Thirty QSAR models were constructed using molecular fingerprints, 2D descriptors, and five machine learning algorithms.
• Structure-activity landscapes (SALs) were visualized using network-like similarity graphs (NSGs), and local discontinuity scores were calculated.

Main Results:

• Local discontinuity scores derived from NSGs positively correlated with cross-validation prediction errors across various models and datasets.
• Innovative ADs, based on pairwise compound similarities, demonstrated superior performance compared to conventional ADs.
• The developed QSAR models and curated datasets provide valuable resources for evaluating PPARγ-related adverse effects.

Conclusions:

• The study successfully developed and validated QSAR models for PPARγ binding affinity prediction.
• Novel SAL analysis and AD definition methods improve the understanding and reliability of QSAR predictions.
• These findings contribute to the efficient and accurate assessment of chemical safety regarding endocrine disruption.