As part of its contribution to the ELIXIR Tools Platform (see also Activities), the node leads the FAIRification of benchmarking datasets for two ELIXIR communities: the 3D-BioInfo Community's PPI Benchmark and the Single Cell Omics Community's BE1 benchmark. Both efforts transform existing benchmark resources into FAIR-compliant, machine-learning-ready datasets with structured metadata, standardised schemas and reproducible evaluation pipelines.
PPI Benchmark FAIRification — 3D-BioInfo Community
ELIXIR 3D-BioInfo PPI Benchmark FAIRification
This project transforms the ELIXIR 3D-BioInfo Benchmark for Protein–Protein Interfaces (1,677 protein interfaces: 844 physiological and 833 non-physiological) into a FAIR-compliant, machine-learning-ready resource, reaching an overall 87.5% FAIR score. The work includes structured JSON-LD metadata with Bioschemas, Schema.org, EDAM and Gene Ontology annotations, packaging via MLCommons Croissant 1.0, structural feature extraction from PDB/mmCIF files, and a complete machine-learning pipeline using cluster-aware GroupKFold cross-validation to prevent data leakage.
| FAIR dimension | Score |
|---|---|
| Findable | 85/100 |
| Accessible | 80/100 |
| Interoperable | 90/100 |
| Reusable | 95/100 |
Implementing the PPI Benchmark as an OpenEBench Experiment
Building on the FAIRification work above, the node is implementing the 3D-BioInfo PPI Benchmark as an OpenEBench-compliant benchmarking experiment, packaging the 14 participant scoring methods, the consensus random-forest predictor and the homology-aware cross-validation splits into the OpenEBench Scientific Benchmarking Data Model (Community, Benchmarking Event, Challenge and Dataset instances). An interactive results/leaderboard viewer, built from this package, is available at tools/oeb_3dbioinfo_ppi_benchmark.
BE1 Benchmark FAIRification — Single Cell Omics Community
ELIXIR Single Cell Omics Benchmark FAIRification
This project FAIRifies the BE1 benchmark dataset from the ELIXIR Single Cell Omics Implementation Study (WP3): a 10x Genomics scRNA-seq experiment with controlled cancer heterogeneity across seven lung cancer cell lines carrying distinct driver mutations (EGFR Del19, KRAS G12S, MET Del14, BRAF G469A, ERBB2 V842I, ROS1 fusion, ALK fusion) plus a PBMC control, totalling 29,528 sequenced cells across 8 samples. The dataset reaches an overall 80.0% FAIR score, with full transformation to the Single Cell Schema v0.4 (16 standardised metadata sheets), persistent identifiers (GEO, DOI, SRA, BioProject, figshare), and an ML-ready benchmark design including an R Shiny app for controlled mixtures and CCLE bulk-transcriptome comparison.
| FAIR dimension | Score |
|---|---|
| Findable | 80/100 |
| Accessible | 80/100 |
| Interoperable | 75/100 |
| Reusable | 85/100 |
Why this matters for the Tools Platform
Both projects follow the same principle: a widely used community benchmark is re-packaged with standardised, machine-actionable metadata (Bioschemas/JSON-LD, Croissant, Single Cell Schema) and validated computational pipelines, so that other groups across ELIXIR can discover, access and reuse these resources to develop and benchmark new computational methods — directly supporting the Tools Platform's goal of an interoperable, well-documented service and data ecosystem.