| KEYWORD |
Benchmarking of tools for protein-ligand binding energy estimation
keywords BINDING FREE ENERGY CALCULATION, DOCKING, DRUG DISCOVERY, PROTEIN-LIGAND INTERACTIONS
Reference persons JACEK ADAM TUSZYNSKI
External reference persons Prof. Maral Aminpour, Department of Biomedical Engineering, University of Alberta, Edmonton, Canada
Research Groups 28- biomedica
Thesis type COMPUTATIONAL
Description Background
Accurately estimating the binding energy between small molecules and biological macromolecules is essential for drug discovery and understanding molecular interactions in biological systems. Computational methods offer powerful tools to approximate binding affinities, often saving time and resources compared to experimental approaches. Among these, molecular docking is widely used for its efficiency, while more detailed approaches like Molecular Mechanics with Generalized Born Surface Area (MM/GBSA) and Poisson–Boltzmann Surface Area (MM/PBSA) offer improved accuracy by combining molecular mechanics energies with continuum solvation models. These methods are particularly popular in research for estimating ligand-binding free energies and have proven useful in both virtual screening and post-docking refinement.
MM/GBSA and MM/PBSA, which typically involve molecular dynamics simulations of the receptor-ligand complex, provide a balance between computational expense and accuracy. They do not require a training set, and their modularity allows for straightforward application to diverse biological targets. However, they rely on approximations that can limit their accuracy in some scenarios. For instance, MM/GBSA and MM/PBSA methods do not always account for the conformational entropy of the binding complex or the presence and thermodynamic contribution of water molecules in the binding site, factors that are crucial for precise energy predictions. Furthermore, various implementations and parameter choices for these methods can lead to inconsistent performance across different systems. [1]
Given these limitations, we seek to assess alternative approaches that might provide better accuracy and possibly reduce computational costs.
Objectives
This project aims to systematically benchmark and compare a range of computational tools for binding energy estimation. By identifying the strengths and weaknesses of each method, the study will offer insights into more accurate, efficient alternatives for computational binding energy estimation in molecular modeling.
Several tools have been developed to automate and refine binding free energy (BFE) calculations, enhancing their accessibility and application in high-throughput settings.
A few are listed below to provide a starting point for the literature review.
● BAT.py [2] integrates with the AMBER simulation package and demonstrates reliability in re-ranking docked poses and estimating absolute binding free energies, with a focus on affordability through graphical processing unit (GPU) optimization.
● GXLE [3] combines MM/GBSA with machine learning (ML) techniques to enhance BFE prediction accuracy. By leveraging ML-based scoring functions to correct errors in MM/GBSA calculations, this hybrid approach achieves improved performance in ranking ligand binding affinities, demonstrating good transferability across diverse protein-ligand systems.
● dPaCS-MD/MSM [4] integrates dissociation Parallel Cascade Selection Molecular Dynamics simulations with Markov state models (MSM) to simulate dissociation pathways and calculate BFE. This method generates accurate BFE profiles and provides mechanistic insights into ligand dissociation, making it valuable for detailed studies on protein-ligand dynamics.
● Advances in end-state methods have also been introduced, such as deepQM [5], which combines linear interaction energy (LIE) methods with ANI-2x neural network potentials to achieve quantum mechanics (QM)-level accuracy at a reduced computational cost. This technique is effective in estimating BFEs across ligands with varied scaffolds, and it achieves a high correlation with experimental values, positioning it as a valuable tool for diverse ligand sets.
● For enhanced sampling, umbrella sampling with additional restraints [6] offers a purely physics-driven approach for BFE estimation
Required skills bioinformatics, chemioinformatics, docking, molecular dynamics
Notes This project will be co-supervised with Prof. Aminpour at the University of Alberta and can be carried out either in Italy or Canada or partly in Italy and partly in Canada
Deadline 26/11/2025
PROPONI LA TUA CANDIDATURA