QSAR modeling of human serum protein binding with several modeling techniques utilizing structure-information...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Pharmacokinetics

Background:

• Human serum protein binding is crucial for drug efficacy and distribution.
• Quantitative Structure-Activity Relationship (QSAR) modeling aids in predicting drug properties.
• A large, curated dataset is essential for robust QSAR model development.

Purpose of the Study:

• To develop and validate QSAR models for predicting human serum protein binding.
• To compare the performance of multiple modeling techniques.
• To identify key molecular descriptors influencing protein binding.

Main Methods:

• Utilized a dataset of 1008 experimental human serum protein binding values.
• Employed four modeling techniques: Multiple Linear Regression (MLR), Artificial Neural Networks (ANN), k-Nearest Neighbors (kNN), and Support Vector Machines (SVM).
• Partitioned data into training (808 compounds) and external validation (200 compounds) sets using structure-based clustering.

Main Results:

• ANN models achieved the highest training set correlation (r²=0.90) and validation set correlation (r²=0.70).
• Ensemble models (ANN, kNN, SVM) generally outperformed MLR.
• Key structure descriptors influencing binding were identified and analyzed.

Conclusions:

• ANN models provide accurate predictions of human serum protein binding.
• The identified structure-activity relationships can guide drug design and modification.
• This work establishes a benchmark dataset for future QSAR studies in protein binding.