Cheminformatic models to predict binding affinities to human serum albumin

Area of Science:

• Medicinal Chemistry
• Pharmacokinetics
• Computational Chemistry

Background:

• Accurate prediction of drug binding to human serum albumin (HSA) is vital for pharmaceutical development, particularly for optimizing pharmacokinetics.
• Existing models may lack broad applicability across diverse chemical spaces.

Purpose of the Study:

• To develop robust quantitative structure-activity relationship (QSAR) models for predicting drug binding affinities to HSA.
• To establish models applicable to a wide range of medicinal chemical compounds.

Main Methods:

• Experimental determination of HSA binding affinities for 95 diverse drugs and druglike compounds using high-performance affinity chromatography.
• Derivation of QSAR models using a genetic algorithm for exhaustive descriptor selection, including multivariate and nonlinear equations.
• Validation of models through internal (cross-validation, randomization) and external testing.

Main Results:

• Developed specific linear models for compound families (e.g., beta-adrenergic antagonists, steroids) with high goodness-of-fit (r² ≥ 0.80).
• Generated global nonlinear models with good fits (r² ≥ 0.78) and strong predictive power (q² ≥ 0.73, external r² ≥ 0.82).
• Identified hydrophobicity (ClogP) as the primary determinant of HSA binding, with structural factors (topological indices, Jurs descriptors) also playing a role.

Conclusions:

• The developed QSAR models provide reliable predictions of HSA binding affinities.
• Drug binding to HSA is governed by a combination of hydrophobic interactions and shape-dependent factors.
• These models can accelerate the design of new drug candidates with improved pharmacokinetic profiles.