MF-SuP-pKa

子图层面的图网络预测pKa，使用了数据增强和迁移学习，原文：MF-SuP-pKa: Multi-fidelity modeling with subgraph pooling mechanism for pKa prediction，代码 MF-SuP-pKa，原文框架如下：

• MF_SuP_pka_model.py 使用了其他文件写的函数，模型代码是 MY_GNN.py 中的 SuP_pka_Predictor
• build_dataset.py 将 smiles 转化成分子图，最后调用 built_data_and_save_for_pka 输出
• build_pka_graph_dataset.py 实际利用数据运行 built_data_and_save_for_pka 并保存输出

def built_data_and_save_for_pka(
        origin_path='data.csv',
        save_g_attentivefp_path='data_graph.bin',
        smiles_path='data_group.csv',
        task_list_selected=None,
        acid_or_base=None):
    data_origin = pd.read_csv(origin_path)
    data_origin = data_origin.fillna(123456)
    labels_list = [x for x in data_origin.columns if x not in ['smiles', 'SMILES', 'group', 'ChEMBL ID',
                                                               'num_acid_sites',
                                                               'num_base_sites']]  # pka_acidic or pka_basic
    if task_list_selected is not None:
        labels_list = task_list_selected
    smiles_name = 'smiles'
    data_set_gnn = build_dataset_with_idx(dataset_smiles=data_origin, labels_list=labels_list,
                                          smiles_name=smiles_name,
                                          acid_or_base=acid_or_base)
    smiles, g_attentivefp, labels, mask, num_sg = map(list, zip(*data_set_gnn))
    graph_labels = {'labels': torch.tensor(labels),
                    'mask': torch.tensor(mask),
                    'num_sg': torch.tensor(num_sg),
                    }


    smiles_pd = pd.DataFrame(smiles, columns=['smiles'])
    smiles_pd.to_csv(smiles_path, index=False)
    print('Molecules graph is saved!')
    save_graphs(save_g_attentivefp_path, g_attentivefp, graph_labels)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28