Title: Protein–ligand binding free-energy campaign
We aim to quantify relative binding free energies for a congeneric ligand series to a target protein and to elucidate binding-mode stability and hydration mechanisms. Using all-atom molecular dynamics with accurate long-range electrostatics, we generate equilibrated
ensembles for apo and holo systems and apply alchemical relative binding free-energy calculations to obtain ΔΔG values with statistically rigorous confidence intervals. The simulations also characterize conformational responses of the protein and key water or ion
participation in the binding site. Results are validated against experimental thermodynamics and available structural data, and the deliverables include ranked ligands with uncertainties near 1 kcal/mol and mechanistic interpretations that inform go/no-go decisions and guide subsequent design.
We are aiming to screen 100 peptides.
Scientific project description:
Objective and hypotheses
- Rank a congeneric series of ligands by relative binding free energy (ΔΔG) to a target protein and validate binding modes observed in experiments (ITC/SPR, crystallography, cryo-EM).
- Secondary aims: quantify protein conformational shifts upon binding and identify key water/ion participation in the active site.
Software
The simulations will be performed using GROMACS with CUDA acceleration (2025.2 or later).
Project requirements
The project consists of 3 phases.
- System building and preprocessing (mostly CPU)
The preparation is executed interactively using a SLURM allocation or in interactive OpenOnDemand sessions (4 times 2 hours on the smallest allocation on the rome partition).
The memory requirements are below 1.75 GB per CPU
The peptides have 1,000 to 10,000 atoms and the generated configuration file are - Short benchmarks and tuning (CPU+GPU)
Determine the best combination of MPI ranks, OpenMP threads, and GPU offload flags for your node type and system size.
We anticipate 10 test runs for both, the CPU and the GPU partition. The runtime for each test does not exceed 30 minutes.
The CPU tests can be executed on thin rome nodes, as the memory requirement is below 1.75 GB per CPU. - Relaxation runs, NPT ensamble
The relaxation runs are executed on 4 rome nodes (512 CPUs) for all 100 peptides.
Depending on the outcome of the previous step, we employ an hybrid MPI / OpenMPI workflow with optimized parameters "num_tasks" and "cpu_per_tasks".
According to the software documentation and user forum, the division of 128 MPI ranks with 4 OpenMP threads each seems to be promising. - Production molecular dynamics runs
The production runs are executed on A100 GPUs. Based on prior experience and literature research we estimate 24 hours per sample.
The total amount of SBU resources are calculated using the Snellius Resource Calculator. To account for the possibility of inaccurate estimate, slower convergence, or, slow IO, we add a margin of +/- 10% and request
- 60,000 thin CPU SBUs
- 350,000 GPU SBUs
