News

At SLING Days 2025, EPICURE HPC support was highlighted through a real-life case study demonstrating how advanced HPC application support enables researchers to scale complex scientific workflows on EuroHPC systems.

The session “From Concept to HPC Computing: EPICURE and a Real-Life Case Study” featured contributions from Žiga Zebec and Samo Miklavc (IZUM), along with Klara Kropivšek Brumat (University of Nova Gorica and Marie-Curie SMASH Fellow).

Unlocking European-level HPC through EPICURE support

The session began with an overview of HPC Vega, Slovenia’s EuroHPC supercomputer, explaining how researchers can transition from an initial concept to full-scale computational execution.

This context set the stage for demonstrating how EPICURE HPC support bridges the gap between resource availability and effective use, offering second-level support for optimisation, performance analysis, code enabling, and large-scale experiment deployment.

A real-life EPICURE HPC case study in computational biology

The second half of the session illustrated a scientific real case from the EPICURE project, presented by Klara Kropivšek Brumat. Her research focuses on nanobody-antigen interactions, which are crucial for the development of next-generation therapeutics.

The study required a massive effort:

• – 123 antibody-antigen complexes;
• – 4 diffusion-based AI models (including AlphaFold 3, Boltz1, Boltz2, and Chai-1);
• – 5 diffusion samples (runs) per complex per model;
• – Over 91,000 individual structure predictions.

This scale would not have been feasible without a strong HPC support. EPICURE’s Application Support Team assisted throughout:

• – Optimising SLURM job arrays for large-scale parallelisation;
• – Introducing an MSA-reuse strategy that reduced runtime by up to 74%;
• – Monitoring GPU behaviour and energy tracking infrastructure;
• – Deploying Singularity containers across HPC systems;
• – Troubleshooting node failures and data-transfer bottlenecks.

These optimisations enabled the team to identify a computational “sweet spot”: running 10-25 diffusion samples captures nearly all quality gains while significantly reducing cost and energy consumption.

Strengthening collaboration and user enablement

Samo Miklavc also detailed common challenges faced by researchers, from environment setup to workflow scalability, and demonstrated how EPICURE’s tailored support helps overcome these hurdles more reliably.