Forecasting CYP2D6 and CYP3A4 Risk with a Global/Local Fusion Model of CYP450 Inhibition.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Cytochrome P450 (CYP450) enzymes are crucial in drug metabolism.
• Predicting CYP450 inhibition early in drug development is vital to mitigate risks.
• Existing Quantitative Structure-Activity Relationship (QSAR) methods have limitations in forecasting novel compounds.

Purpose of the Study:

• To develop a novel QSAR method for predicting CYP450 inhibition.
• To forecast the potential risks of compounds not yet synthesized.
• To leverage extensive corporate CYP450 inhibition data.

Main Methods:

• A global/local fusion method is employed, creating custom QSAR models on-the-fly for each prediction.
• A consensus approach combines descriptor-based models and pharmacophore fingerprint similarity.
• A chronological data split is used for forward prediction testing and validation.

Main Results:

• The QSAR models achieved high accuracy, with standard errors approaching data limitations for confident predictions (0.4 log IC50 units).
• Classification accuracy for CYP2D6 and CYP3A4 activity reached 79% for predicted actives and 85% for predicted inactives.
• The method demonstrated robust validation through a novel forward prediction scheme.

Conclusions:

• The developed QSAR method effectively utilizes corporate data to predict future compound risks.
• The on-the-fly model customization and consensus approach enhance prediction accuracy and uncertainty assessment.
• This approach offers a valuable tool for early-stage drug safety assessment and risk mitigation.