Biomedical Science

The work carried out in the HANAMI project, focusing on biomedical applications, will investigate molecular simulations and respiratory diseases.
In the case of molecular simulations, this area of research is seen as a complementary tool to experimental techniques. Molecular dynamics simulations provide high space and time resolution and a better understanding of molecular processes. Researchers will work to improve data-driven simulations by integrating different information from different experimental techniques to guide the sampling of states (with the European code GROMACS and the Japanese GENESIS).

In a second step, the scaling of molecular simulations will also be improved – with the integration of the fast-multipole method (FMM) into GROMACS. The world-leading Japanese ExaFMM code is being adapted for molecular and will be linked with GROMACS to enable much better scaling of molecular simulations. For this work, we need access to a diverse set of available HCP architectures when combining European and Japanese resources. It is the hope to understand better the structures and functioning of proteins, which will also allow for a better understanding of how diseases work and, consequently, how the production of medicines should be altered to increase effectiveness and reduce side effects. The second part, the evaluation of respiratory diseases, is currently done mainly through questionnaires filled out by patients or visual inspection of computer tomography (CT) images. Although this is a quick way of obtaining information about respiratory conditions, it is not possible to ascertain the causes of respiratory problems. This is where a three-dimensional analysis of the flow field in the patient’s nasal cavity can make a difference. Recently, the inclusion of Computational Fluid Dynamics (CFD) simulations in decision-making has become more prevalent. CFD simulations can also be extremely useful for estimating hospital infection risks due to exhaled droplets, e.g., for patients in a waiting room. For both types of respiratory flow analysis, CFD simulations must be employed in automated workflows to allow the integration of clinics. To address these challenges, a combined approach of CFD, HPC, and machine learning (ML) will be required from an internal and external perspective (with the European code m-AIA and the Japanese code CUBE). The internal aspect is dedicated to assisting the surgeon in the decision-making process to treat respiratory diseases. An automated surgery planning tool with limitations at this stage must be improved with European and Japanese resources and techniques. The results can be expected to improve success rates of surgeries (in interventions to correct a deviation of the nasal septum, this figure is only 55%). The external aspect will focus on breathing in waiting rooms, including spreading droplets from, e.g., coughing. The goal is to optimize the interior design, e.g., ventilation inlets. From the internal perspective, the efficiency of optimization will be improved by integrating jointly developed ML algorithms.

In the scope of WP5, researchers will work on Project 1 — Exascale electrostatics & machine learning to enable molecular dynamics of cell-size systems.

The strategy of the molecular life science pillar is to extend existing collaborations on software and hardware in molecular dynamics simulation into long-term coordinated development of the two major-impact codes GROMACS (EU) and GENESIS (RIKEN), with additional coordination with the simulation hardware developed in the RIKEN MD-GRAPE project. The two codebases complement each other since GENESIS has focused on pure high-end ARM CPU scaling on Fugaku, while GROMACS is highly tuned and portable across a range of GPU (NVIDIA, AMD, Intel) and CPU architectures, and the strategic collaboration aims to have both codes used as benchmarks and optimization targets in the post-Fugaku projects, which will be a unique co-design opportunity for both the European and RIKEN sides.

Regarding Project 2 — Development of genome analysis pipelines for Personalized Medicine, it will build upon the ongoing collaboration between groups in Japan and Europe in the context of the International Cancer Genome Consortium (ICGC), the follow-up ICGC/TCGA Pan-Cancer Analysis of Whole Genomes (PCAWG) Project and the current ICGC-Argo project. More specifically, part of the project is based on the work already developed by Prof. David Torrents (BSC) during his 2022 sabbatical in Japan. Prof. Torrents collaborated with the mentioned Japanese groups on developing the genome analysis workflows, as these are key components of the proposed project. Additional contacts established during Prof. Barillot’s visit to RIKEN in collaboration with PRAIRIE will be explored.

The scientific goal of this project is to provide mechanistic models for individual patients reflecting their differential genomic composition and response to treatments. These mechanistic models will be interpreted with field experts to identify potential treatments and/or associated signatures. By developing such personalized medicine infrastructure, the proposed work will contribute to the technical development of practical implementations that will facilitate the work of clinicians and genome scientists, ultimately leading to better and more tailored treatment strategies for cancer patients.

Still related to WP5, Project 3 — Personalized Medicine on Macroscopic Level: Respiratory Flows and Infection Risk Analyses, builds upon a long-year collaboration of the Jülich Supercomputing Centre (JSC), Forschungszentrum Jülich, with the RIKEN Center for Computational Science (R-CCS) that takes place within the scope of two joint projects under the umbrella of the Joint Laboratory for Extreme Scale Computing (JLESC). In the first project Comparison of Meshing and CFD Methods for Accurate Flow Simulations on HPC systems, the relevant codes m-AIA and CUBE from RWTH Aachen University and RIKEN R-CCS were ported and benchmarked on the Jülich and RIKEN R-CCS machines. The recent project, Deep Neural Networks for CFD Simulations, includes the development of AI methods for engineering applications with a focus on macroscopic biomedical computations. The m-AIA code and a use case for analysing respiratory flows for personalized medicine are part of CoE RAISE.

The first aim of this project is to develop AI-assisted automated methodologies that enable the running of large-scale computational fluid dynamics (CFD) simulations, including an analysis of the results for individual patients. The second aim is to use AI methods and novel hardware technologies at JSC and RIKEN R-CCS to accelerate computations. Providing an AI-assisted surgery planning tool to surgeons as a demonstrator is the third goal of this project. The methodologies developed to achieve these aforementioned goals contribute to achieving the last aim of this project: assessing the risk of exhaled infectious (virus). All implementations complement missing elements in either the m-AIA or CUBE code.

Keep exploring
our project!

Results

Learn more