Prediction of Drug-Plasma Protein Binding Using Artificial Intelligence Based Algorithms.

Area of Science:

• Pharmacology
• Computational Chemistry
• Drug Discovery

Background:

• Plasma protein binding (PPB) significantly influences drug distribution and clinical development.
• Early consideration of drug distribution properties is essential in drug design.
• Accurate PPB prediction models are increasingly important for pharmaceutical research.

Purpose of the Study:

• To develop and evaluate computational models for predicting plasma protein binding.
• To identify the most effective machine learning algorithm for PPB prediction.
• To provide a tool for screening drug candidates in early development phases.

Main Methods:

• A systematic approach using machine learning algorithms including Support Vector Machine (SVM), Artificial Neural Network (ANN), k-Nearest Neighbor (k-NN), Probabilistic Neural Network (PNN), Partial Least Square (PLS), and Linear Discriminant Analysis (LDA).
• Utilized a diverse dataset of 736 drugs/drug-like compounds with various in vitro and in silico molecular descriptors.
• Compared the performance of different algorithms for predicting plasma protein binding.

Main Results:

• The Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel demonstrated superior performance compared to other algorithms.
• SVM achieved high accuracy (89.73% training, 89.97% validation), precision (92.56%), sensitivity (87.26%), specificity (91.97%), and F1 score (0.898).
• The developed SVM model showed robust predictive capabilities for plasma protein binding.

Conclusions:

• The developed SVM model is a valuable tool for predicting plasma protein binding.
• This predictive model can assist in the preliminary screening of drug candidates.
• The findings support the integration of computational models in early drug design and development to mitigate risks associated with unfavorable PPB.