|
---
|
|
license: mit
|
|
---
|
|
# Description
|
|
Metal Ion Binding prediction is a binary classification task where each input protein *x* is mapped to a label *y* ∈ {0, 1}, corresponding to whether there are metal ion–binding sites in the protein.
|
|
|
|
The digital label means:
|
|
|
|
0: No
|
|
|
|
1: Yes
|
|
|
|
# Splits
|
|
|
|
**Structure type:** PDB
|
|
|
|
The dataset is from [**Exploring evolution-aware & -free protein language models as protein function predictors**](https://arxiv.org/abs/2206.06583). We employ all proteins from the original dataset, and split them based on 70% structure similarity (see [ProteinShake](https://github.com/BorgwardtLab/proteinshake/tree/main)), with the number of training, validation and test set shown below:
|
|
|
|
- Train: 5797
|
|
- Valid: 719
|
|
- Test: 719
|
|
|
|
# Data format
|
|
|
|
We organize all data in LMDB format. The architecture of the databse is like:
|
|
|
|
**length:** The number of samples
|
|
|
|
**0:**
|
|
|
|
- **name:** The PDB ID of the protein
|
|
- **chain:** The chain ID of the protein
|
|
- **seq:** The structure-aware sequence
|
|
- **label:** Digital label of the sequence
|
|
|
|
**1:**
|
|
|
|
**···**
|
|
|
|
|
