Matched Molecular Series Analysis for ADME Property Prediction

Area of Science:

• Medicinal Chemistry
• Computational Drug Design
• Pharmacokinetics

Background:

• Drug design relies heavily on generating and prioritizing new molecules.
• Matched molecular series analysis (MMSA) is a formal approach integrating these crucial design elements.
• Evaluating MMSA's performance as an ADME property prediction tool is essential for understanding its capabilities and limitations.

Purpose of the Study:

• To assess the performance of Matched Molecular Series Analysis (MMSA) as a tool for predicting ADME properties.
• To compare MMSA's predictive accuracy against standard machine learning models and matched molecular pair analysis.
• To identify metrics for estimating the reliability of MMSA predictions, akin to machine learning's applicability domain.

Main Methods:

• Utilized four large, diverse in-house datasets: logD, microsomal clearance, CYP2C9, and CYP3A4 inhibition.
• Applied the MMSA concept of parallel structure-activity relationship (SAR) for property profile similarity transfer between molecular series.
• Tested four similarity metrics, with centered root-mean-square deviation (cRMSD) and a network score combination yielding the best performance.

Main Results:

• The combination of centered root-mean-square deviation (cRMSD) and a network score provided the highest prediction accuracy.
• cRMSD alone offered the optimal balance between predictive accuracy and the number of transferable predictions.
• Statistical metrics were identified to estimate MMSA prediction reliability, analogous to machine learning applicability domains.
• MMSA demonstrated prediction accuracy comparable to standard machine learning and matched molecular pair analysis.
• MMSA's interpretability of prediction origins surpasses that of typical machine learning models.

Conclusions:

• MMSA is a valuable and interpretable tool for ADME property prediction in drug design.
• The method provides insights into SAR transferability and aids in identifying potential data errors.
• MMSA offers a viable alternative or complement to machine learning approaches for molecule prioritization.