Automation of absolute protein-ligand binding free energy calculations for docking refinement and compound evaluation

Germano Heinzelmann, Michael K. GilsonScientific 2021 · DOI

TL;DR

This paper introduces BAT.py, a Python software package that automates complex absolute binding free energy (ABFE) calculations for protein-ligand systems. By leveraging GPU acceleration and supporting multiple ABFE methods, BAT.py aims to make these powerful simulations more accessible for early-stage drug discovery, enabling faster and more cost-effective compound evaluation and pose refinement.

What Problem Does This Paper Address?

Estimating the binding affinity of drug candidates to target proteins is crucial in drug discovery. Absolute binding free energy (ABFE) calculations using molecular simulations offer a promising way to achieve this, potentially reducing the need for expensive and time-consuming experimental screening. However, performing these calculations is notoriously complex, requiring significant expertise in setting up simulations, assigning force field parameters, and analyzing results. This complexity has been a major bottleneck, limiting the widespread adoption of ABFE methods, especially for screening diverse compound libraries or refining poses generated by docking algorithms. The authors aim to bridge this gap by providing an automated, user-friendly tool that simplifies and accelerates the ABFE workflow.

Methods

The core of this work is the BAT.py software, a Python package designed to automate ABFE calculations using the AMBER simulation suite. BAT.py supports three primary ABFE methodologies:

  • Double Decoupling (DD): A standard alchemical method where the ligand is gradually decoupled from the protein and solvent. This method can encounter numerical artifacts with charged ligands due to changes in system charge.
  • Attach-Pull-Release (APR): A method that simulates the physical unbinding of the ligand. While avoiding charge issues, it can be challenging for ligands in buried binding sites.
  • Simultaneous Decoupling-Recoupling (SDR): A novel variant of DD that decouples the ligand from the binding site while simultaneously recoupling it to bulk solvent. This approach maintains a constant system charge, mitigating artifacts associated with charged ligands and making it suitable for both neutral and charged species, even with limited solvent access.

BAT.py automates several key steps, including:

  • Ligand protonation and parameter assignment.
  • System setup, solvation, and ion addition.
  • Application of protein and ligand restraints to control conformational freedom and improve simulation stability.
  • Generation and execution of free energy windows.
  • Calculation of binding free energies using TI/MBAR.

The software leverages GPU acceleration through AMBER's pmemd.cuda for significant speedups, making calculations more computationally feasible.

Datasets

The paper evaluates BAT.py using two protein-ligand systems:

  • BRD4 bromodomain (PDB ID 5uf0): A system with a charge-neutral ligand (2-methyl-5-(methylamino)-6-phenylpyridazin-3(2H)-one, referred to as 89J). Calculations were performed for the crystal pose and five poses generated by AutoDock Vina.
  • MCL-1 protein (PDB ID 4hw2): A system with a charged ligand (6-chloro-3-[3-(4-chloro-3,5-dimethylphenoxy)propyl]-1H-indole-2-carboxylic acid, referred to as 19H). Calculations were performed for the crystal pose and five poses generated by AutoDock Vina.

The docked poses were generated using a modified version of the CELPP tutorial workflow, employing AutoDock Vina for docking and UCSF Chimera for system preparation. The structural similarity of docked poses to the crystal structure was assessed using Root-Mean-Square Deviation (RMSD).

Key Insights and Ideas

  • Automation is Key: The primary insight is that automating the complex ABFE workflow is crucial for making these methods practical for drug discovery. BAT.py demonstrates that this automation is achievable and significantly reduces the barrier to entry.
  • SDR Method's Advantage: The introduction and successful application of the SDR method highlight its utility for handling charged ligands, a common challenge in drug discovery, by avoiding numerical artifacts.
  • GPU Acceleration for Throughput: The reliance on GPU computing demonstrates a clear path towards achieving high-throughput ABFE calculations, making it feasible to screen larger libraries of compounds.
  • Pose Refinement Power: The results show that ABFE calculations can effectively distinguish between accurate and inaccurate docked poses, providing a valuable tool for refining docking predictions.
  • Cost-Effectiveness: The combination of automation and GPU usage suggests that high-performance ABFE calculations can be performed at a relatively low cost, further enhancing their applicability.

Key Results

  • BRD4 System: The calculated binding free energy for the crystallographic pose of 89J to BRD4 was -6.1 kcal/mol, showing good agreement with the experimental value of -5.2 kcal/mol. Crucially, docked poses with lower RMSD values (≤ 2.0 Å) yielded binding free energies consistent with the crystal structure and within 1.5 kcal/mol of experiment. Inaccurate poses (RMSD ≥ 5.4 Å) resulted in significantly less favorable binding free energies.
  • MCL-1 System: For the charged ligand 19H binding to MCL-1, the SDR method yielded a binding free energy of -12.2 kcal/mol for the crystal pose, compared to an experimental value of -10.0 kcal/mol. Similar to the BRD4 system, docked poses with low RMSD (< 2.0 Å) showed similar computed binding free energies to the crystal pose, while less accurate poses resulted in weaker binding free energies.
  • Computational Performance: The study reports that a single GTX 1070 GPU can perform a complete binding free energy calculation for a single pose in approximately 6.64 days. The authors project that with newer GPUs and full parallelization, this time could be reduced to around 3 hours per ligand.
  • Method Comparison: Comparisons between MBAR and TI-GQ methods for calculating free energy components showed good consistency, with MBAR generally reporting lower uncertainties. The SDR method also showed good agreement with the conventional DD method for a neutral ligand.

Limitations and Future Work

  • Limited Test Cases: The authors acknowledge that the presented results are based on only two protein-ligand systems and serve as an initial proof of principle. More extensive testing across diverse protein systems, force fields, and water models is needed.
  • Pose Generation Dependence: The accuracy of the ABFE calculations is still dependent on the quality of the initial ligand poses generated by docking programs.
  • Future Directions: The authors plan to extend their work to other protein systems and explore different force fields and simulation parameters. They also aim to investigate accelerated sampling techniques and hybrid Monte Carlo/MD methods to further improve convergence and efficiency. The ultimate goal is to establish ABFE calculations as a routine tool for high-throughput virtual screening in early-stage drug discovery.

Summary updated

Abstract

Absolute binding free energy calculations with explicit solvent molecular simulations can provide estimates of protein-ligand affinities, and thus reduce the time and costs needed to find new drug candidates. However, these calculations can be complex to implement and perform. Here, we introduce the software BAT.py, a Python tool that invokes the AMBER simulation package to automate the calculation of binding free energies for a protein with a series of ligands. The software supports the attach-pull-release (APR) and double decoupling (DD) binding free energy methods, as well as the simultaneous decoupling-recoupling (SDR) method, a variant of double decoupling that avoids numerical artifacts associated with charged ligands. We report encouraging initial test applications of this software both to re-rank docked poses and to estimate overall binding free energies. We also show that it is practical to carry out these calculations cheaply by using graphical processing units in common machines that can be built for this purpose. The combination of automation and low cost positions this procedure to be applied in a relatively high-throughput mode and thus stands to enable new applications in early-stage drug discovery.
BibTeX
@inproceedings{heinzelmann2021automation,
  author = {Germano Heinzelmann and Michael K. Gilson},
  title = {{Automation of absolute protein-ligand binding free energy calculations for docking refinement and compound evaluation}},
  booktitle = {Scientific},
  year = {2021},
  doi = {10.1038/s41598-020-80769-1},
}
Figure