Exploring the Upper Boundaries of Distance-Based Protein-Ligand Binding Affinity Prediction: Accuracy Improvements via Linear Regression and Artificial Neural Networks

Optimized power exponents for distance-based variables improve protein-ligand binding affinity prediction, with neural network models reaching R values between 0.800 and 0.900.

• Why it matters: Accurate prediction of protein-ligand binding affinity is crucial for computational drug discovery, but traditional fixed-form scoring functions limit predictive flexibility and accuracy.
• What they did: The study integrated optimized distance-dependent descriptors with linear regression and artificial neural network algorithms, training and validating models on the PDBbind 2024 dataset with thousands of complexes.
• The result: Neural network hybrid models significantly outperformed previous scoring functions, achieving high accuracy close to experimental error, demonstrating that rationally-optimized exponents can enhance predictive performance.