Empirical Scoring Functions for Affinity Prediction of Protein-ligand Complexes.

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Accurate assessment of protein-ligand complex quality and binding affinity is fundamental for computer-aided drug design (CADD).
• Simple filtering suffices for early-stage virtual screening, but refined docking and hit optimization require precise binding free energy estimates.
• Rigorous free energy calculations are often impractical, necessitating the use of scoring functions for affinity estimation.

Purpose of the Study:

• To evaluate the role and effectiveness of empirical scoring functions in estimating protein-ligand binding affinity for CADD.
• To explore the application of machine learning methods in developing improved scoring functions.
• To identify current limitations and future challenges in scoring function development.

Main Methods:

• Empirical scoring functions were developed using a regression-based approach.
• Training data comprised experimental structures and affinity data of protein-ligand complexes.
• Descriptors capturing essential interaction features were employed.
• Both classical linear regression and machine learning techniques were utilized for training.

Main Results:

• Machine learning methods demonstrated considerable improvements in prediction accuracy on large, generic datasets.
• Empirical scoring functions provide practical affinity estimates when rigorous calculations are infeasible.
• Despite advancements, existing scoring functions still present limitations.

Conclusions:

• Empirical scoring functions, particularly those enhanced by machine learning, are valuable tools for estimating binding affinity in CADD.
• Further research is needed to overcome existing limitations and enhance the predictive power of scoring functions.
• Continued development is essential for advancing virtual screening and drug optimization processes.