EOSC Data Commons Granted EOSC EU Node Credits

The benchmark datasets used in this project are inherently large in scale, each requiring multiple rounds of docking, scoring, and clustering. Running the full benchmark suite across all categories and all scenarios would require a significant amount of compute time and resources. Due to these large compute requirements, it was not feasible to run the complete benchmarks on the EOSC EU Node infrastructure within the allocated compute resources. Instead, the resources were used to carry out targeted test runs on a smaller, representative subset of the data. The primary objectives of these test runs were to:

• Validate that the full workflow pipeline was correctly set up and functional on the EOSC EU Node environment.
• Confirm that all required dependencies, tools, and configurations were properly resolved.
• Verify that the execution of individual pipeline steps – including docking, clustering and scoring- was completed successfully end-to-end.
• Identify any environment-specific issues or incompatibilities that would need to be addressed before scaling up.

These test runs were successful and provided confidence that the workflows are reproducible. The full benchmarks can be executed in subsequent phases on partner-provided or in-house HPC infrastructure.

Each benchmark category was structured to evaluate multiple docking protocols and parameter configurations, allowing a systematic comparison of different approaches. For instance, for the protein–protein benchmark, five distinct scenarios were defined, each encoded in a dedicated YAML configuration file:

These scenarios are designed to cover a spectrum from default protocol runs – which provide a useful baseline – through to more specialised configurations, including ab initio docking and alternative clustering strategies. The inclusion of both HADDOCK 2.4 and HADDOCK 3 scenarios enables direct performance comparison between the two versions of the software.

Analogous scenario structures were prepared for the remaining benchmark categories (protein–ligand, protein–peptide, protein–DNA, and protein–glycan), ensuring a consistent and comparable evaluation framework across all benchmark datasets covered in the project.

Biological systems are complex – proteins interact with DNA, RNA, sugars, peptides, and small molecules in ways that underpin virtually every cellular process and disease mechanism. Integrative modelling platforms like HADDOCK3 allow researchers to model these interactions at scale by combining experimental data with computational approaches across diverse scenarios. Additionally, this work delivers a comprehensive resource: curated benchmark datasets spanning protein–protein, protein–DNA, protein–RNA, protein–peptide, protein–glycan, and protein–ligand systems, advanced docking scenarios tailored to each system, and an automated tool capable of executing multiple docking runs simultaneously and reproducibly — on both HPC clusters and cloud infrastructure.