Alchemical free energy calculations, commonly used in computer-aided drug design, is inherently slower than standard molecular dynamics (MD) simulations. Access to massive HPC resources, with GPUs, means that a large number of calculations can be run in parallel. It is important to utilise both CPU and GPU resources as efficiently as possible with a proper distribution of workload.
The LIGATE (LIgand Generator and portable drug discovery platform AT Exascale) EU project has improved the performance of GROMACS free-energy simulations significantly. For version 2022, collaborations between the BioExcel CoE and LIGATE helped to extend this SIMD support also to the non-bonded free-energy kernels, and the performance was improved further for version 2024 (see Figure). This development can improve the performance of free-energy calculations up to 300%, which has major implications for cost efficiency of large-scale simulations on supercomputers. On top of that, the GPU support for free-energy simulations was enhanced. Since version 2021, PME calculations can be offloaded to the GPU even if charges are modified by an alchemical transformation.
The LIGATE workflow and project work packages.
High-throughput virtual screening is an important tool for drug development. Commonly, primitive docking scoring functions are used to estimate binding affinities. MD simulations provide more accurate prediction of the binding free energy, but the throughput is relatively low and preparing the calculations, including parameterising the ligands for biomolecular force fields, is a hurdle. By automating the workflow, LIGATE has facilitated large-scale MD simulations on HPC resources. The collaboration between BioExcel and LIGATE has significantly improved the hardware utilisation and shortened the calculation time. The improvements by BioExcel to GROMACS resulted in significant speed-ups of alchemical free energy calculations. In turn, these improvements mean that more protein-ligand complexes can be simulated per day and better hardware utilisation implies power savings.
