---
license: cc-by-4.0
language:
- en
pretty_name: ProtGlycanDock-Minimal
size_categories:
- 1K<n<10K
tags:
- biology
- protein
- glycan
- docking
- benchmark
configs:
- config_name: default
  data_files:
  - split: train
    path: metadata/splits/train.csv
  - split: valid
    path: metadata/splits/valid.csv
  - split: protein_holdout_test
    path: metadata/splits/protein_holdout_test.csv
  - split: glycan_holdout_test
    path: metadata/splits/glycan_holdout_test.csv
---

# ProtGlycanDock Dataset – Minimal Version

> **NOTE**: This is the **minimal version** of ProtGlycanDock, uploaded for the NeurIPS 2026 Datasets and Benchmarks Track due to the < 4 GB size requirement. It contains only the mmCIF structural files and the metadata tables needed to define splits and case information.  
> The full dataset (including PDB structures, JSON inputs, and reconstruction scripts) is available in the accompanying complete repository (https://huggingface.co/datasets/ProtGlycanDock/ProtGlycanDock).

ProtGlycanDock is a curated protein–glycan docking dataset. This minimal release includes the mmCIF files for all **1,781** protein–glycan complexes, along with split assignments and case-level metadata.

| Subset               | Split label           | Cases | Purpose                                                    |
| -------------------- | --------------------- | ----: | ---------------------------------------------------------- |
| Train                | `train`               |  1374 | Training set used for model fitting.                       |
| Validation           | `valid`               |   125 | Validation set for model selection.                        |
| Protein-holdout test | `test`                |   148 | Evaluation set constructed from held-out protein clusters. |
| Glycan-holdout test  | `glycan_holdout_test` |   134 | Evaluation set constructed from held-out glycans.          |

**NOTE**: JSON input files and processed PDB structures are only available in the full dataset release.



## Repository Layout

The repository is organized as follows:

```text
data/
  mmcif/
    {PDB_ID}.cif          		# mmCIF file for each complex

metadata/
  dataset_file_manifest.csv		# File manifest for the full dataset (for reference)
  splits/
  	train.csv
  	valid.csv
  	protein_holdout_test.csv
  	glycan_holdout_test.csv