Exploring additional computational resources for Latvian genome reference project

• Name: Exploring additional computational resources for Latvian genome reference project
• EuroHPC machine used: MeluXina
• Topic: Biological Sciences

Overview of the project

This project has sequenced ~4000 human whole genomes for the Latvian genome reference and started secondary analyses on Latvia’s largest HPC (RTU). Because local resources are limited for timely completion, the project benchmarked and optimised their containerised WGS variant-calling workflow (nf-core/sarek + Nextflow) on EuroHPC resources (MeluXina), including multi-node execution and GPU acceleration, to identify scalable configurations suitable for large-scale production.

How did EPICURE support the project and what were the benefits of the support?

The Exploring additional computational resources for Latvian genome reference project requested support to adapt and benchmark their existing containerised Nextflow/Sarek WGS pipeline on MeluXina in a way that (1) avoids single-node limitations, (2) prevents inefficient submission of many small scheduler jobs with long queue times, and (3) enables GPU-accelerated tools correctly. EPICURE helped them reconfigure execution to use the HyperQueue executor on pre-allocated nodes, apply MeluXina best practices for data placement, and troubleshoot/enable the GPU configuration (including resolving minor dependency issues needed for Parabricks-enabled runs).

Turnaround time / throughput: Best configuration (GPU nodes + Parabricks + HyperQueue) reduced the benchmark runtime from ~14.6 h (initial CPU-only setup) to ~4.73 h (3 GPU nodes), i.e. about a 3.1× speed-up.
Scalability know-how: Demonstrated practical multi-node scaling using HyperQueue and clarified that biggest gains come from GPU-aware tools, not just running on GPU nodes.
Operational reliability: Produced a “known-good” configuration and best practices that can be reused for large-scale Latvian genome processing.