High-resolution crystal structures leverage protein binding affinity predictions.

Area of Science:

• Computational biology
• Structural biology
• Biophysics

Background:

• Predicting protein binding affinities from structural data is challenging due to diverse binding modes.
• Existing methods either oversimplify models with few variables or risk overfitting with numerous parameters.

Purpose of the Study:

• To develop improved models for predicting protein binding affinities using structural information.
• To identify key variables that enhance prediction accuracy and understand factors influencing prediction difficulty.

Main Methods:

• Utilized the Structure-Affinity-Benchmark (SAB) dataset with 144 cases.
• Developed models using 12 variables for enthalpic and entropic changes upon binding.
• Employed a model selection procedure to identify optimal sparse models from these variables.
• Introduced new parameters encoding interface morphology and packing properties.

Main Results:

• Achieved significant improvements in affinity prediction accuracy.
• Models predict dissociation constant (Kd) within 1 and 2 orders of magnitude for 48% and 79% of cases, respectively.
• Performance increased to 62% and 89% for high-resolution structures.
• New parameters related to interface morphology and packing were crucial for performance.
• Interface flexibility did not correlate with prediction difficulty; flexible cases showed comparable performance.

Conclusions:

• The developed models offer a marked improvement in predicting protein binding affinities.
• Interface morphology and packing properties are key determinants of binding affinity.
• High-resolution structural data and carefully selected variables can enhance prediction accuracy.
• The findings suggest that affinity prediction is feasible with appropriate datasets and modeling strategies.