---
license: apache-2.0
language:
- en
tags:
- Protein Langauge Model
- AI for Drug Discovery
- AI for Science
---

# Dataset Summary

This is a protein structure and protein prediction dataset for the papers GearNet (https://arxiv.org/abs/2203.06125), ESM-GearNet (https://arxiv.org/abs/2303.06275), ESM-S (https://arxiv.org/abs/2402.05856) and ProtIR ().
The dataset is first processed by DeepFRI (https://www.nature.com/articles/s41467-021-23303-9) and then collected by CDConv (https://openreview.net/forum?id=P5Z-Zl9XJ7). 
The original files can also be downloaded from the github of CDConv (https://github.com/hehefan/Continuous-Discrete-Convolution).
We upload the dataset here for the ease of downloading.

The details are shown in the following table.

| Dataset          | #Train | #Valid | #Test 95%      | #Test 50%           | #Test 30%  |
|------------------|--------|--------|----------------|---------------------|------------|
| EnzymeCommission | 15,550 | 1,729  | 1,919          | 1,117               | 720        |
| GeneOntology     | 29,898 | 3,322  | 3,416          | 2,199               | 1,717      |
| Fold             | 12,312 | 736    | 1,272 (Family) | 1,254 (Superfamily) | 718 (Fold) |

# Evaluation Pipeline
Please refer to ESM-S (https://github.com/DeepGraphLearning/esm-s) for the training and evaluation code.