---
license: unknown
task_categories:
- tabular-classification
- graph-ml
- text-classification
tags:
- chemistry
- biology
- medical
pretty_name: MoleculeNet Tox21
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: train
    path: "tox21.csv"
---

# MoleculeNet Tox21

Tox21 dataset [[1]](#1), part of MoleculeNet [[2]](#2) benchmark. It is intended to be used through 
[scikit-fingerprints](https://github.com/scikit-fingerprints/scikit-fingerprints) library.

The task is to predict 12 toxicity targets, including nuclear receptors and stress response pathways. All tasks are binary.

Note that targets have missing values. Algorithms should be evaluated only on present labels. For training data, you may want to impute them, e.g. with zeros.

| **Characteristic** |      **Description**     |
|:------------------:|:------------------------:|
|        Tasks       |            12            |
|      Task type     | multitask classification |
|    Total samples   |           7831           |
|  Recommended split |         scaffold         |
| Recommended metric |           AUROC          |

**Warning:** in newer RDKit vesions, 8 molecules from the original dataset are not read correctly due to disallowed
hypervalent states of their aluminium atoms (see [release notes](https://github.com/rdkit/rdkit/releases/tag/Release_2024_09_1)).
This version of the Tox21 dataset contains manual fixes for those molecules, removing additional hydrogens, e.g. `[AlH3] -> [Al]`.
In OGB scaffold split, used for benchmarking, only the first 1 of those problematic 8 is from the test set. Applied mapping is:
```
"NC(=O)NC1N=C(O[AlH3](O)O)NC1=O" -> "NC(=O)NC1N=C(O[Al](O)O)NC1=O"
"O=CO[AlH3](OC=O)OC=O" -> "O=CO[Al](OC=O)OC=O"
"CC(=O)O[AlH3](O)O" -> "CC(=O)O[Al](O)O"
"CC(=O)O[AlH3](O)OC(C)=O" -> "CC(=O)O[Al](O)OC(C)=O"
"CCOC(=O)/C=C(/C)O[AlH3](OC(C)CC)OC(C)CC" -> "CCOC(=O)/C=C(/C)O[Al](OC(C)CC)OC(C)CC"
"CCCCO[AlH3](OCCCC)OCCCC" -> "CCCCO[Al](OCCCC)OCCCC"
"O=S(=O)(OC[C@H]1O[C@H](O[C@]2(COS(=O)(=O)O[AlH3](O)O)O[C@H](COS(=O)(=O)O[AlH3](O)O)[C@@H](OS(=O)(=O)O[AlH3](O)O)[C@@H]2OS(=O)(=O)O[AlH3](O)O)[C@H](OS(=O)(=O)O[AlH3](O)O)[C@@H](OS(=O)(=O)O[AlH3](O)O)[C@@H]1OS(=O)(=O)O[AlH3](O)O)O[AlH3](O)O.O[AlH3](O)[AlH3](O)O.O[AlH3](O)[AlH3](O)O.O[AlH3](O)[AlH3](O)O.O[AlH3](O)[AlH3](O)O" -> "O=S(=O)(OC[C@H]1O[C@H](O[C@]2(COS(=O)(=O)O[Al](O)O)O[C@H](COS(=O)(=O)O[Al](O)O)[C@@H](OS(=O)(=O)O[Al](O)O)[C@@H]2OS(=O)(=O)O[Al](O)O)[C@H](OS(=O)(=O)O[Al](O)O)[C@@H](OS(=O)(=O)O[Al](O)O)[C@@H]1OS(=O)(=O)O[Al](O)O)O[Al](O)O.O[Al](O)[Al](O)O.O[Al](O)[Al](O)O.O[Al](O)[Al](O)O.O[Al](O)[Al](O)O"
"CCCCCCCCCCCCCCCCCC(=O)O[AlH3](O)O" -> "CCCCCCCCCCCCCCCCCC(=O)O[Al](O)O"
```

## References
<a id="1">[1]</a> 
Tox21 Challenge
https://tripod.nih.gov/tox21/challenge/

<a id="2">[2]</a> 
Wu, Zhenqin, et al.
"MoleculeNet: a benchmark for molecular machine learning."
Chemical Science 9.2 (2018): 513-530
https://pubs.rsc.org/en/content/articlelanding/2018/sc/c7sc02664a