Read-Across Structure-Property Relationship-Based Superior Prediction of Fraction Unbound in Plasma from Chemical...

Area of Science:

• Medicinal Chemistry
• Pharmacokinetics
• Computational Chemistry

Background:

• Accurate prediction of fraction unbound in plasma (f_up) is vital for early-stage drug discovery.
• Traditional methods often use complex, opaque models requiring extensive descriptors.
• Minimizing late-stage failures and refining screening processes necessitates reliable f_up prediction.

Purpose of the Study:

• To develop interpretable models for predicting f_up using a read-across strategy combined with quantitative structure-property relationships (QSPR).
• To minimize descriptor complexity while maintaining high predictive performance.
• To provide insights into structure-property relationships governing plasma protein binding.

Main Methods:

• Application of the read-across strategy integrated with traditional QSPR.
• Development of interpretable regression and classification models.
• Validation and comparison against various machine learning methods.

Main Results:

• Quantitative read-across structure-property relationship multiple linear regression and support vector classifier models demonstrated superior predictive performance.
• The developed models showed high accuracy across diverse chemical compounds.
• The approach successfully minimized descriptor complexity, enhancing model interpretability.

Conclusions:

• The read-across QSPR strategy offers a powerful and interpretable approach for predicting f_up.
• This method aids in understanding the relationship between chemical structures and plasma protein binding.
• The findings contribute to more rational drug design and the development of effective therapeutics.