BioExcel-2 D1.5 – Long-term hardware and application roadmap

This deliverable reports updates on the BioExcel-2 core application roadmaps as presented in deliverable D1.3. The roadmaps have been updated for work performed, as well as for changes in needs driven by both developments on the hardware side and requests of users. In addition, there have been changes in prioritization and new tasks added due to the outbreak of Covid-19, which has spurred a massive amount of research in life sciences. This, in combination with the no-cost extension of the BioExcel-2 project duration, has led to some development tasks being moved to the extension period. The GROMACS development has focused on a number of key user-requested features, in particular adding multiple time stepping, driving simulations with external data, facilitating generation of input data, and better free energy tools. Some of this is still ongoing and a few new user requests have been identified that are worth implementing. The development of performance optimization has prioritized porting of the GPU acceleration of GROMACS to SYCL, targeting AMD GPUs for the new LUMI pre-exascale machine. Also multi-GPU PME is nearly ready through co-design with NVIDIA. Focus on these urgent tasks has meant work on other tasks started later. For the medium term, future effort is planned on performant and flexible APIs to enable more flexible ensemble parallelism for the Exascale. The development of the PMX package followed two main directions. Firstly, the core code was updated with the changes necessary for the successful future usability of the software. In particular, considering the depreciation of Python2, pmx was rewritten in Python3. The continuous support of the main pmx functionalities was ensured. The second major development direction comprises the incorporation of features frequently requested by users, the addition of new mutation libraries and creation of workflows for free energy calculation setup and analysis. HADDOCK’s development focuses on the continuous improvement of the web server, which functions as the main user’s interface, the implementation of relevant features and the design of the new version HADDOCK3. The web server is being co-developed alongside the local HADDOCK version, so that new features (coarse-grained and shape docking, etc.) are readily available in an intuitive and responsive user-friendly interface. A major rework of the current code is in place for what will become HADDOCK3, with a modular design we aim to provide flexibility for workflow creation and a straightforward framework to integrate new tools. This new modular architecture will also enable more efficient execution on HPC resources. To provide users with access to chemical reactivity in biomolecular simulations, we will develop an interface between GROMACS and CP2K, a highly parallelized density functional theory (DFT) computer program. With the new interface, users can perform multi-scale simulations, in which one part of the system is modelled with DFT, while the remainder is described with a molecular mechanics force field (i.e. QM/MM). Our developments focus on providing the community with an accurate, scalable and user-friendly QM/MM solution. Compared to the previous BioExcel software roadmap, which focused on CPMD for DFT functionality in QM/MM simulations, this updated roadmap reflects the project’s transition to CP2K, which has more wide-spread availability and uptake in the communit