| # Flexynesis Benchmark Datasets |
|
|
| | Dataset key | Biology | Modalities | Samples | Task types available | |
| |---|---|---|---|---| |
| | `dataset1` | Cancer drug response (CCLE/GDSC cell lines) | `gex`, `cnv` | ~950 / 240 | Regression (drug IC50 per compound) | |
| | `dataset2` | Microsatellite instability (MSI) | `gex`, `meth` | ~380 / 95 | Binary classification (MSI-H vs MSS) | |
| | `lgggbm_tcga_pub_processed` | Brain tumours: LGG + GBM (TCGA) | `mut`, `cna` | 556 / 238 | Classification, survival, regression | |
| | `brca_metabric` | Breast cancer (METABRIC) | `gex`, `cna` | ~1390 / 595 | Classification, survival, regression | |
| | `singlecell_bonemarrow` | Bone marrow single-cell RNA | `gex` | ~7500 / 2500 | Classification, unsupervised | |
|
|
| **Note on single-cell data:** flexynesis was designed for bulk multi-omics data (patient cohorts, cell lines) — not single-cell RNA-seq. It has no built-in handling for the sparsity, scale, or batch structure typical of scRNA-seq. The `singlecell_bonemarrow` dataset is included as a benchmark curiosity and works well for **supervised cell type classification** (where cell type labels are available), but flexynesis is not the right tool for unsupervised single-cell analysis, trajectory inference, or integration of large scRNA-seq atlases. For those tasks, use Scanpy/Seurat/scVI instead. If the user's question is specifically about supervised classification of single-cell data with known labels, it is worth trying. |
|
|
| All datasets hosted at https://bimsbstatic.mdc-berlin.de/akalin/buyar/flexynesis-benchmark-datasets/ |
|
|
| Reference: Uyar et al., *Nature Communications* 2025 — https://doi.org/10.1038/s41467-025-63688-5 |
|
|
