https://github.com/psi-cmd/synprepy
Single step retrosynthesis prediction
https://github.com/psi-cmd/synprepy
Last synced: 12 months ago
JSON representation
Single step retrosynthesis prediction
- Host: GitHub
- URL: https://github.com/psi-cmd/synprepy
- Owner: psi-cmd
- Created: 2025-06-26T11:52:46.000Z (12 months ago)
- Default Branch: main
- Last Pushed: 2025-06-30T04:32:38.000Z (12 months ago)
- Last Synced: 2025-06-30T05:28:21.918Z (12 months ago)
- Language: Jupyter Notebook
- Size: 79.1 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 化学反应逆向预测
## Python 版本
python3.6
## 安装依赖
* RDKit
```bash
# 安装rdkit
conda install -c conda-forge rdkit
# 安装pytorch
请按照官网安装: https://pytorch.org/
# 安装pyg
请按照官网安装: https://pytorch-geometric.readthedocs.io/en/latest/notes/installation.html
# 安装其他依赖
cd synprepy # 进入项目文件夹
pip install -r requirements.txt
```
## 准备数据
1. 准备具有如下反应数据的tsv文件
| rid | rxn_smi |
| --- | ------- |
| 0 | [OH:2][N:3]1[N:4]=[N:5][C:6]2=[C:7]1[CH:8]=[CH:9][CH:10]=[CH:11]2>>[OH:2][N:3]1[N:4]=[N:5][C:6]2=[C:7]1[CH:8]=[CH:9][CH:10]=[CH:11]2 |
| 1 | [O:13]1[CH2:14][CH2:15][CH2:16][CH2:17]1>>[O:13]([CH2:14][C@@H:15]1[O:13][C@H:17]2[C@@H:15]([CH2:14][CH2:16]1)[CH2:16]2)[CH3:17] |
## 填写配置文件（config.json）
```json
{
"root": "存储反应数据和反应模板数据的文件夹路径",
"rxn_data_tsv_file_name": "反应数据文件名（该文件必须处于上个设置的root文件夹内）",
"rid_with_rxn_template_tsv_file_name": "保持默认即可",
"rxn_centralized_template_tsv_file_name": "保持默认即可",
"rxn_extended_template_tsv_file_name": "保持默认即可",
"rid_with_tid_tsv_file_name": "保持默认即可",
"rxn_centralized_template_selector_config": {
"root": "存储中心反应模板选择器的相关文件的文件夹",
"train_rids_file_name": "保持默认即可",
"test_rids_file_name": "保持默认即可",
"train_temp_dir_name": "保持默认即可",
"test_temp_dir_name": "保持默认即可",
"min_num_covered_rxns_by_rxn_centralized_template": "过滤模板时要求模板的最少覆盖反应数",
"filter_tids_file_name": "保持默认即可",
"device": "cpu或者cuda",
"batch_size": 1000,
"epoch_num": 200,
"lr_start": 0.0005,
"lr_end": 1e-7
},
"min_num_covered_rxns_by_rxn_extended_template": 50
}
```
## 单步反应预测
1. 抽取反应模板：
运行 rxn_template_prepare.py
2. 训练单步反应模板选择器：
运行 rxn_template_selector/rct_train_test.py