Improving the Stability and Transferability of Effective ADMET Models by Adding Quantum Mechanical Descriptors

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Discovery

Background:

• Accurate prediction of ADMET properties is crucial for drug development.
• Traditional structural descriptors have limitations in predicting complex molecular behaviors.

Purpose of the Study:

• To evaluate the impact of combining quantum mechanical (QM) descriptors with structural descriptors for ADMET property prediction.
• To assess the performance of QM descriptors in both in-space and out-of-space prediction scenarios.

Main Methods:

• Incorporated QM descriptors alongside traditional structural descriptors.
• Evaluated prediction accuracy for human liver microsomes (HLM) stability, permeability, solubility, and hERG inhibition.
• Analyzed feature importance and assessed extrapolation capabilities.

Main Results:

• QM descriptors enhanced prediction accuracy across all tested ADMET endpoints.
• Significant improvements were observed for permeability, with QM dispersion energy being a key feature.
• QM descriptors demonstrated superior extrapolation capabilities and stability in reduced data scenarios.

Conclusions:

• QM descriptors offer more comprehensive insights than structural descriptors alone.
• A hybrid approach combining QM and structural descriptors improves predictive accuracy and robustness.
• QM descriptors are particularly valuable for challenging extrapolation tasks and small datasets in ADMET prediction.