Do All Paths Lead to Rome? How Reliable is Umbrella Sampling Along a Single Path?

Area of Science:

• Computational chemistry
• Biophysics
• Molecular modeling

Background:

• Molecular dynamics (MD) simulations are crucial for estimating binding free energies of molecular complexes.
• Umbrella sampling (US) is a common MD technique for calculating potential of mean force (PMF) along a reaction coordinate.
• Reproducibility and convergence of US for binding free energy calculations remain under-investigated.

Purpose of the Study:

• To systematically assess the reproducibility and convergence of the standard umbrella sampling (US) workflow.
• To compare the performance of US with adaptive biasing methods, specifically accelerated weight histogram (AWH), for binding free energy calculations.
• To highlight potential pitfalls in established US protocols and advocate for improved methods.

Main Methods:

• Investigated reproducibility of US along a distance-based reaction coordinate for protein-ligand and protein-protein complexes.
• Performed complete US workflow repetitions to quantify discrepancies in binding free energies.
• Employed accelerated weight histogram (AWH) simulations to assess convergence with simultaneous multi-pathway sampling.

Main Results:

• Standard US workflows exhibited significant variability (2-20 kcal/mol) in computed binding free energies due to pathway differences.
• Small variations in binding pathways lead to unavoidable discrepancies in US results.
• Accelerated weight histogram (AWH) simulations yielded consistent and reproducible PMF profiles across tested systems.

Conclusions:

• Established umbrella sampling protocols may suffer from poor reproducibility in binding free energy calculations.
• Adaptive biasing methods like AWH demonstrate superior consistency and reliability for these calculations.
• Further benchmarking of adaptive biasing methods is recommended for accurate binding affinity predictions.