QSAR Development for Plasma Protein Binding: Influence of the Ionization State

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Pharmacokinetics

Background:

• Plasma protein binding (PPB) is a critical factor influencing drug efficacy and disposition.
• Accurate prediction of PPB is essential for drug discovery and development.
• Existing quantitative structure-activity relationship (QSAR) models often require improvement for reliable PPB prediction.

Purpose of the Study:

• To explore strategies for enhancing the performance of QSAR models for plasma protein binding (PPB).
• To investigate the impact of endpoint transformation, chemical representation, data consistency, and applicability domain definition on model accuracy.
• To develop robust QSAR models for predicting human fraction unbound (Fu).

Main Methods:

• Retrieved and curated human fraction unbound (Fu) data for 670 compounds from literature.
• Calculated molecular descriptors considering ionization state at physiological pH (7.4).
• Employed various algorithms and chemical descriptors, including SMILES-based string descriptors with CORAL software.
• Performed outlier analysis to define model applicability domains.

Main Results:

• Achieved Q² values close to 0.60 in internal validation for selected models.
• External validation yielded r² values consistently greater than 0.60.
• The CORAL descriptor-based model for the square root of Fu (√fu) showed the best performance with an external validation r² of 0.74.

Conclusions:

• The developed QSAR models demonstrated robustness and suitability for real-world applications, including chemical screening for ADMET profiling.
• Optimizing descriptors, particularly considering ionization states, is crucial for accurate prediction of PPB for ionized molecules.