---
license: cc-by-4.0
---

## 1. Files

### gnn_cnn_data.zip 

The dataset used to train gnn and cnn-3d model. 

### split_60.zip

The dataset used to train Uni-Mol or ProFSA model for the SIU 0.6 version. It contains train.lmdb, valid.lmdb and test.lmdb.

### split_90.zip

The dataset used to train Uni-Mol or ProFSA model for the SIU 0.9 version. It contains train.lmdb, valid.lmdb and test.lmdb.

### pretrain_weights.zip

Pretrained weights for Uni-Mol and ProFSA.

profsa.pt: weights for pretrained ProFSA model

mol_pre_no_h_220816.pt: weights for pretrained Uni-Mol molecular Encoder

pocket_pre_220816.pt: weights for pretrained Uni-Mol pocketr Encoder

### final_dic.pkl

Complete dataset file in pickle format

### result_structures.zip

contains all structure files for protein and docked small molecules.



## 2. Dataset Format

### final_dic.pkl

each key is an uniprot id

corresponding value is a list of dictionaries. Each dictionary is a data point and has following keys:

| Key      | Description     |
|----------------|----------------|
| atoms  | atom types in ligand  |
| coordinates |  list of different conformations of the ligand |
|  pocket_atoms| atom types in pocket |
|  pocket_coordinates | atom positions of the pocket |
|  source_data |  pdbid and uniprot id information |
|  label |  dictionary for different labels |
|  ik |  inchikey of ligand |
|  smi |  smiles of ligand |



### Other lmdb files

All data are in lmdb format, and have the same keys as shown above.

Note that for single task learning, the albel is a float value instead of a dictionary.